+ All Categories
Home > Documents > Feynman checkerboard - turgor.ru

Feynman checkerboard - turgor.ru

Date post: 14-Feb-2022
Category:
Upload: others
View: 11 times
Download: 0 times
Share this document with a friend
35
0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.00002 0.00004 0.00006 0.00008 0.00010 βˆ™ βˆ™ )) // ** // oo OO OO // oo 44
Transcript

Feynman checkerboard

An intro to algorithmic quantum field theory

E. Akhmedova, M. Skopenkov, A. Ustinov, R. Valieva, A. Voropaev

0.001

0.002

0.003

0.004

0.005

0.006

0.007

Basic model from Β§1

0.00002

0.00004

0.00006

0.00008

0.00010

Model with mass from Β§3

Figure 1: The probability tofind an electron at a given point(white depicts large oscillations)

Summary. It turns out that rainbow patterns on soapbubbles andunbelievable laws of motion of electrons can be explained by means ofone simple-minded model. It is a game, in which a checker moves on acheckerboard by certain simple rules, and we count the turnings. Thisβ€˜Feynman checkerboard’ can explain all phenomena in the world (witha serious proviso) except atomic nuclei and gravitation. We are goingto solve mathematical problems related to the game and discuss theirphysical meaning; no knowledge of physics is assumed.

Main results. The main results are explicit formulae for

βˆ™ the percentage of light of given color reflected by a glass plate ofgiven width (Problem 24);

βˆ™ the probability to find an electron at a given point, if it was emit-ted from the origin (Problem 15; see Figure 1).

Although the results are stated in physical terms, they are mathemati-cal theorems, because we provide mathematical models of the physicalphenomena in question, with all the objects defined rigorously. Moreprecisely, a sequence of models with increasing precision.

Plan. We start with a basic model and upgrade it step by stepin each subsequent section. Before each upgrade, we summarize whichphysical question does it address, which simplifying assumptions doesit resolve or impose additionally, and which experimental results doesit explain. Our aim is what is called 2-dimensional quantum electro-dynamics but the last steps on this way (sketched in Sections 8–10)still have not been done. (A 4-dimensional one can already explain allphenomena β€” with proviso and exceptions β€” but we do not discuss it.)

The scheme of upgrades dependence might help to choose your way:

1. Basic model

))

6. External field // 8***. Interaction

**

4. Source

οΏ½οΏ½

2. Spin //oo

οΏ½οΏ½

OO

7. Identical particles & antiparticles

οΏ½οΏ½

OO

10***. QED

5. Medium 3. Mass //oo 9***. Creation & annihilation

44

Conventions. If a problem is a statement of an assertion, then it is requested to prove the assertion.A puzzle is a problem, in which both a precise statement and a proof are requested. Hard problems aremarked with stars; you get first rank in Feynman checkerboard for solving one, and masters in Feynmancheckerboard for solving three from three different sections. Solutions are accepted in writing but you mayspend an earned star for 10 attempts to submit a solution in oral form (successful or not). If you cannotsolve a problem, proceed to the next ones: they may provide hints. Even if you do not reach the top(the proofs of main results), you learn much. You are encouraged to state and try to prove also your ownobservations and conjectures; you become a grand-master in Feynman checkerboard for discovering a newnontrivial one (and maybe even write your own scientific paper).

1

1 Basic modelQuestion: what is the probability to find an electron at the point (π‘₯, 𝑦), if it was emitted from the point (0, 0)?Assumptions: no self-interaction, no creation of electron-positron pairs, unit mass and lattice step, point source;no nuclear forces, no gravitation, electron moves uniformly along the 𝑦-axis and does not move along the 𝑧-axis.Results: double-slit experiment, charge conservation.

Figure 2: Checker paths

On an infinite checkerboard, a checker moves to the diagonal-neighboring squares, either upwards-right or upwards-left. To each path𝑠 of the checker, assign a vector οΏ½οΏ½(𝑠) as follows. Start with a vector oflength 1 directed upwards. While the checker moves straightly, the vectoris not changed, but each time when the checker changes the direction, thevector is rotated through 90∘ clockwise (independently of the direction thechecker turns). In addition, at the very end the vector is divided by 2(π‘¦βˆ’1)/2,where 𝑦 is the total number of moves. The final position of the vector iswhat we denote by οΏ½οΏ½(𝑠). For instance, for the path in Figure 2 to the top,the vector οΏ½οΏ½(𝑠) = (1/8, 0) is directed to the right and has length 1/8.

Denote οΏ½οΏ½(π‘₯, 𝑦) :=βˆ‘

𝑠 οΏ½οΏ½(𝑠), where the sum is over all the paths of thechecker from the square (0, 0) to the square (π‘₯, 𝑦), starting with the upwards-right move. Set οΏ½οΏ½(π‘₯, 𝑦) := 0, if there are no such paths. For instance,οΏ½οΏ½(1, 3) = (0,βˆ’1/2)+(1/2, 0) = (1/2,βˆ’1/2). The length square of the vectorοΏ½οΏ½(π‘₯, 𝑦) is called the probability1 to find an electron in the square (π‘₯, 𝑦), if itwas emitted from the square (0, 0). Notation: 𝑃 (π‘₯, 𝑦) := |π‘Ž(π‘₯, 𝑦)|2.

In Figure 1 to the top, the color of a point (π‘₯, 𝑦) with even π‘₯+𝑦 depictsthe value 𝑃 (π‘₯, 𝑦). The sides of the apparent angle are not the lines 𝑦 = Β±π‘₯ (and nobody knows why!).

In what follows squares (π‘₯, 𝑦) with even and odd π‘₯ + 𝑦 are called black and white respectively.

1. Observations for small 𝑦. Answer the following questions for each 𝑦 = 1, 2, 3, 4 (and state your ownquestions and conjectures for arbitrary 𝑦): Find the vector οΏ½οΏ½(π‘₯, 𝑦) and the probability 𝑃 (π‘₯, 𝑦) for each π‘₯.When 𝑃 (π‘₯, 𝑦) = 0? What is

βˆ‘π‘₯∈Z 𝑃 (π‘₯, 𝑦) for fixed 𝑦? What are the directions of οΏ½οΏ½(1, 𝑦) and οΏ½οΏ½(0, 𝑦)?

The probability2 to find an electron in the square (π‘₯, 𝑦) subject to absorption in the square (π‘₯β€², 𝑦′) isdefined analogously to 𝑃 (π‘₯, 𝑦), only the summation is over those paths 𝑠 that do not pass through (π‘₯β€², 𝑦′).The probability is denoted by 𝑃 (π‘₯, 𝑦 bypass π‘₯β€², 𝑦′).

-1000 -500 500 1000

-0.05

0.05

Figure 3: π‘Ž2(π‘₯, 1000)2. Double-slit experiment. Is it true that 𝑃 (π‘₯, 𝑦) = 𝑃 (π‘₯, 𝑦 bypass 0, 2)+𝑃 (π‘₯, 𝑦 bypass 2, 2)? Is it true that 𝑃 (π‘₯, 𝑦) β‰₯ 𝑃 (π‘₯, 𝑦 bypass π‘₯β€², 𝑦′)?

3. Find 𝑃 (0, 12). How to table the values οΏ½οΏ½(π‘₯, 𝑦) quickly without exhaustionof all paths? (The first solution grants first rank in Feynman checkers.)

Denote by π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) the coordinates of οΏ½οΏ½(π‘₯, 𝑦); see Figure 3.

4. Dirac’s equation. Express π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) through π‘Ž1(π‘₯Β± 1, 𝑦 βˆ’ 1) and π‘Ž2(π‘₯Β± 1, 𝑦 βˆ’ 1).

5. Probability/charge conservation. For each positive integer 𝑦 we haveβˆ‘

π‘₯∈Z 𝑃 (π‘₯, 𝑦) = 1.

6. Symmetry. How the values π‘Ž1(π‘₯, 100) for π‘₯ < 0 and for π‘₯ β‰₯ 0 are related with each other? The samefor the values π‘Ž1(π‘₯, 100) + π‘Ž2(π‘₯, 100).

7. Huygens’ principle. What is a fast way to find οΏ½οΏ½(π‘₯, 199), if we know οΏ½οΏ½(π‘₯, 100) for all integers π‘₯?

8. Using a computer, plot the graphs of the functions 𝑓𝑦(π‘₯) = 𝑃 (π‘₯, 𝑦) for various 𝑦, joining each pair ofpoints (π‘₯, 𝑓𝑦(π‘₯)) and (π‘₯ + 2, 𝑓𝑦(π‘₯ + 2)) by a segment; cf. Figure 3. The same for the function π‘Ž1(π‘₯, 𝑦).

9.* Find an explicit formula for the vector οΏ½οΏ½(π‘₯, 𝑦) and the probability 𝑃 (π‘₯, 𝑦) (it is allowed to use a sumwith at most 𝑦 summands in the answer).

10.** (Skipable puzzle) Guess a simple β€œapproximate formula” for οΏ½οΏ½(π‘₯, 𝑦) and 𝑃 (π‘₯, 𝑦), accurate for |π‘₯| β‰ͺ 𝑦.1One should think of the value 𝑦 as fixed, and the squares (βˆ’π‘¦, 𝑦), (βˆ’π‘¦ + 2, 𝑦), . . . , (𝑦, 𝑦) as all the possible outcomes of

an experiment. For instance, the 𝑦-th horizontal might be a photoplate detecting the electron.Familiarity with probability theory is not required for solving the presented problems.Beware that our rule for the probability computation is valid only for the basic model in question; we are going to changethe rule slightly in the upgrades. We make similar remarks each time we break some fundamental principles for simplicity.

2Thus an additional outcome of the experiment is that the electron has been absorbed and has not reached the photoplate.

2

2 Spin

Question: what is the probability to find a right electron at (π‘₯, 𝑦), if a right electron was emitted from (0, 0)?Assumptions: the same.Results: spin reversal.

The trick in the solution of the previous problems has a physical meaning: it is convenient to consideran electron as being in one of the two states: right-moving or left-moving. We write just β€˜right ’ or β€˜left ’for brevity3. This is not just a convenience but reflects an inalienable electron’s property called spin4.

Denote οΏ½οΏ½(π‘₯, 𝑦,+) :=βˆ‘

𝑠 οΏ½οΏ½(𝑠), where the sum is over only those paths from (0, 0) to (π‘₯, 𝑦), which bothstart and finish with an upwards-right move. Define οΏ½οΏ½(π‘₯, 𝑦,βˆ’) to be an analogous sum over paths whichstart with an upwards-right move but finish with an upwards-left move.

The length square of the vector οΏ½οΏ½(π‘₯, 𝑦,+) (respectively, οΏ½οΏ½(π‘₯, 𝑦,βˆ’)) is called the probability5 to find aright (respectively, left) electron in the square (π‘₯, 𝑦), if a right electron was emitted from the square (0, 0).Denote by 𝑃 (π‘₯, 𝑦,+) := |π‘Ž(π‘₯, 𝑦,+)|2 and 𝑃 (π‘₯, 𝑦,βˆ’) := |π‘Ž(π‘₯, 𝑦,βˆ’)|2 these probabilities.

11. Express οΏ½οΏ½(π‘₯, 𝑦,+) and οΏ½οΏ½(π‘₯, 𝑦,βˆ’) through π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦); 𝑃 (π‘₯, 𝑦) through 𝑃 (π‘₯, 𝑦,+) and 𝑃 (π‘₯, 𝑦,βˆ’).

12.* Spin reversal. What is the probability 𝑃 (𝑦0,βˆ’) :=βˆ‘

π‘₯∈Z 𝑃 (π‘₯, 𝑦0,βˆ’) to find a left electron on theline 𝑦 = 𝑦0 (it is allowed to use a sum with at most 𝑦0 summands in the answer)? Find the maximalelement and the limit of the sequence 𝑃 (1,βˆ’), 𝑃 (2,βˆ’), 𝑃 (3,βˆ’), . . . .

A crash-course in calculus. For this and other problems with stars, the following formuli might be useful.Recall that

βˆ‘π‘›π‘˜=0

12π‘˜

= 2βˆ’ 12𝑛 . Clearly, as 𝑛 increases, the sum becomes closer and closer to 2. We would like

to writeβˆ‘βˆž

π‘˜=012π‘˜

= 2; let us give a definition of such an infinite sum. A sequence π‘Ž1, π‘Ž2, π‘Ž3, . . . has a limit π‘Ž, iffor each real πœ€ > 0 there is 𝑁 such that for each integer 𝑛 > 𝑁 we have |π‘Žπ‘› βˆ’ π‘Ž| < πœ€. Notation: π‘Ž = limπ‘›β†’βˆž π‘Žπ‘›.For instance, limπ‘›β†’βˆž

(2βˆ’ 1

2𝑛

)= 2. By definition, put

βˆ‘βˆžπ‘˜=0 π‘Žπ‘˜ = limπ‘›β†’βˆž

βˆ‘π‘›π‘˜=0 π‘Žπ‘˜. Then indeed

βˆ‘βˆžπ‘˜=0

12π‘˜

= 2.The following generalization is called Newton’s binomial theorem (allowed to use without proof):

(1 + π‘₯)π‘Ÿ =

βˆžβˆ‘π‘˜=0

π‘Ÿ(π‘Ÿ βˆ’ 1) Β· Β· Β· (π‘Ÿ βˆ’ π‘˜ + 1)

π‘˜(π‘˜ βˆ’ 1) Β· Β· Β· 1π‘₯π‘˜

for each complex π‘₯ with |π‘₯| < 1 and each real π‘Ÿ, or for π‘₯ = 1 and π‘Ÿ > βˆ’1. In particular, for π‘Ÿ = βˆ’1 and βˆ’12 we get

1

1βˆ’ π‘₯=

βˆžβˆ‘π‘˜=0

π‘₯π‘˜ and1√1βˆ’ π‘₯

=βˆžβˆ‘π‘˜=0

2π‘˜(2π‘˜ βˆ’ 1) Β· Β· Β· (π‘˜ + 1)

π‘˜(π‘˜ βˆ’ 1) Β· Β· Β· 1π‘₯π‘˜

4π‘˜.

The following Stirling formula allows to estimate the summands in Newton’s binomial theorem:

√2πœ‹ π‘˜π‘˜+1/2π‘’βˆ’π‘˜ ≀ π‘˜(π‘˜ βˆ’ 1) Β· Β· Β· 1 ≀ 𝑒 π‘˜π‘˜+1/2π‘’βˆ’π‘˜.

Here 𝑒 denotes limπ‘›β†’βˆž(1 + 1/𝑛)𝑛. It is an irrational number between 2.71 and 2.72.

3 MassQuestion: what is the probability to find a right electron of mass π‘š at (π‘₯, 𝑦), if it was emitted from (0, 0)?Assumptions: the mass and the lattice step are now arbitrary.Results: a formula for the probability for small lattice step.

To check our model against experiment we need the following generalization.Fix πœ€,π‘š > 0 called lattice step and particle mass respectively. To each path 𝑠 of the checker, assign

a vector οΏ½οΏ½(𝑠,π‘šπœ€) as follows. Start with the vector (0, 1). While the checker moves straightly, the vectoris not changed, but each time when the checker changes the direction, the vector is rotated through 90∘

3Beware that in 3 or more dimensions β€˜right ’ and β€˜left ’ mean something very different from the movement direction.Although often visualized as the direction of the electron rotation, these states cannot be explained in nonquantum terms.

4And chirality ; beware that the term spin usually refers to a property, not related to the movement direction at all.5Thus an experiment outcome is a pair (final π‘₯-coordinate, last-move direction), whereas the final 𝑦-coordinate is fixed.

These are the fundamental probabilities, whereas 𝑃 (π‘₯, 𝑦) should in general be defined by the formula from the solution ofProblem 11 rather than the above formula 𝑃 (π‘₯, 𝑦) = |π‘Ž(π‘₯, 𝑦)|2 (being a coincidence).

3

clockwise and multiplied by π‘šπœ€. In addition, at the very end the vector is divided by (1 + π‘š2πœ€2)(π‘¦βˆ’1)/2,where 𝑦 is the total number of moves. The final position of the vector is what we denote by οΏ½οΏ½(𝑠,π‘šπœ€). Thevectors οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€,Β±) and the numbers 𝑃 (π‘₯, 𝑦,π‘šπœ€,Β±) are defined analogously to οΏ½οΏ½(π‘₯, 𝑦,Β±) and 𝑃 (π‘₯, 𝑦,Β±),only οΏ½οΏ½(𝑠) is replaced by οΏ½οΏ½(𝑠,π‘šπœ€). For instance, 𝑃 (π‘₯, 𝑦, 1,+) = 𝑃 (π‘₯, 𝑦,+). In Figure 1 to the bottom, thecolor of a point (π‘₯, 𝑦) with even π‘₯ + 𝑦 depicts the value 𝑃 (π‘₯, 𝑦, 0.02,+) + 𝑃 (π‘₯, 𝑦, 0.02,βˆ’).

13. (Puzzle) Massless and heavy particles. Find 𝑃 (π‘₯, 𝑦, 0,+) and define 𝑃 (π‘₯, 𝑦,∞,+) for each π‘₯, 𝑦.

14. Solve analogues of Problems 4, 5, and 9 for π‘šπœ€ = 1.

In an experiment, we measure the probabilities to find the electron in intervals π‘₯0 ≀ π‘₯ ≀ π‘₯0 + βˆ†π‘₯,𝑦 = 𝑦0 rather than at particular points. Here π‘₯0, 𝑦0,βˆ†π‘₯ are not integers but actual lengths measured inmeters. If all squares have small size 1

𝑛× 1

𝑛, then the interval is approximated by the collection of black

squares(2βŒŠπ‘›π‘₯2

βŒ‹, 2

βŒŠπ‘›π‘¦2

βŒ‹)with π‘₯, 𝑦 satisfying the above (in)equalities. This leads to the following problem.6

15.* (First main problem) Continuum limit. For each π‘₯, 𝑦,π‘š find limπ‘›β†’βˆž 𝑛 οΏ½οΏ½(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹, π‘šπ‘›,βˆ’

)and

limπ‘›β†’βˆž 𝑛 οΏ½οΏ½(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹, π‘šπ‘›,+

)In the answer, it is allowed to use the following expressions7

𝐽0(𝑧) :=βˆžβˆ‘π‘˜=0

(βˆ’1)π‘˜(𝑧/2)2π‘˜

(π‘˜!)2and 𝐽1(𝑧) :=

βˆžβˆ‘π‘˜=0

(βˆ’1)π‘˜(𝑧/2)2π‘˜+1

π‘˜!(π‘˜ + 1)!.

4 SourceQuestion: what is the probability to find a right electron at (π‘₯, 𝑦), if it was emitted by a source of wave length πœ†?Assumptions: the source is now realistic.Results: wave propagation.

Figure 4: Checker paths maynow start at distinct squares

A realistic source does not produce electrons localized at π‘₯ = 0 (asin our game) but a rather wide wave impulse instead. For our game,this means that the checker can start from an arbitrary black square onthe horizontal line 𝑦 = 0 (not too far from the origin), but the initialdirection of the vector οΏ½οΏ½(𝑠) is rotated through an angle proportional tothe distance from the starting square to the origin; see Figure 4.

Formally, fix real πœ€, πœ† > 0 and odd βˆ† called lattice step, wave length,and impulse width respectively. Denote by 𝑅𝛼 οΏ½οΏ½ the rotation of a vector οΏ½οΏ½through the angle |𝛼|, which is counterclockwise for 𝛼 β‰₯ 0 and clockwisefor 𝛼 < 0. Define the vector

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,+) :=1βˆšβˆ†

Ξ”βˆ’1βˆ‘π‘₯0=1βˆ’Ξ”π‘₯0 even

βˆ‘π‘ 

𝑅2πœ‹π‘₯0πœ€/πœ† οΏ½οΏ½(𝑠),

where the second sum is over all checker paths 𝑠 from the square (π‘₯0, 0) to the square (π‘₯, 𝑦), starting andending with an upwards-right move. The length square of the vector is the probability to find a right electronat (π‘₯, 𝑦), emitted by a source of wave length πœ† and impulse width βˆ†. It is denoted by 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ†,+).Define οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,βˆ’) and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ†,βˆ’) analogously. For instance, 𝑃 (π‘₯, 𝑦, πœ†/πœ€, 1,+) = 𝑃 (π‘₯, 𝑦,+)for each πœ†, πœ€, and οΏ½οΏ½(π‘₯ + 1, 1, πœ†/πœ€,βˆ†,+) = 1√

Ξ”

(βˆ’ sin 2πœ‹π‘₯πœ€

πœ†, cos 2πœ‹π‘₯πœ€

πœ†

)for even |π‘₯| < βˆ†.

16. Let βˆ† = 3, πœ†/πœ€ = 4. Find the vector οΏ½οΏ½(π‘₯, 𝑦, 4, 3,+) and the probability 𝑃 (π‘₯, 𝑦, 4, 3,+) for 𝑦 = 1, 2, 3and each π‘₯. What is

βˆ‘π‘₯∈Z(𝑃 (π‘₯, 𝑦, 4, 3,+)+𝑃 (π‘₯, 𝑦, 4, 3,βˆ’)) for fixed 𝑦 = 1, 2, 3? When 𝑃 (π‘₯, 3, 4, 3,+) = 0?

17. Probability/charge conservation. Solve analogues of Problems 4 and 5 for οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,βˆ’),οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,+), and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ†,+) + 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ†,βˆ’) instead of π‘Ž1(π‘₯, 𝑦), π‘Ž2(π‘₯, 𝑦), and 𝑃 (π‘₯, 𝑦).

18. Causality. Both√

βˆ†οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,+) and βˆ†π‘ƒ (π‘₯, 𝑦, πœ†/πœ€,βˆ†,+) do not depend on βˆ† for βˆ† > 𝑦 + |π‘₯|.Denote8 οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Β±)=

βˆšβˆ†οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ†,Β±) and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,Β±)=βˆ†π‘ƒ (π‘₯, 𝑦, πœ†/πœ€,βˆ†,Β±) for any βˆ†>𝑦+|π‘₯|.

19. Wave. How to find οΏ½οΏ½(π‘₯, 100, πœ†/πœ€,+) for all even π‘₯, if we know it for just one even π‘₯?

20.* Wave propagation. For each π‘₯, 𝑦, πœ†, πœ€ find οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’), 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’), 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’)+𝑃 (π‘₯, 𝑦, πœ†/πœ€,+).6In the limits, the normalization factor of 𝑛 is a bit harder to explain; we not discuss it.7Called Bessel functions, which are almost as well-studied as sine and cosine; but familiarity with them is not required.8The notation οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Β±) should not be confused with οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€,Β±).

4

5 MediumQuestion: which part of light of given color reflects from a glass plate of given width?Assumptions: right angle of incidence, no polarization of light; the mass now depends on π‘₯ but not on the color.Results: thin-film reflection.

Our model can be applied also to describe propagation of light in transparent media such as glass9.Light propagates as if it had some nonzero mass inside the media, while the mass remains zero outside10.The wavelength determines the color of light.

21. (Puzzle) Define an analogue of οΏ½οΏ½(𝑠,π‘šπœ€) in the case when the mass π‘š = π‘š(π‘₯) depends on π‘₯ so thatanalogues of Problems 4 and 5 remain true.

Given a mass π‘š = π‘š(π‘₯), define οΏ½οΏ½(π‘₯, 𝑦,π‘š(π‘₯)πœ€, πœ†/πœ€,Β±) and 𝑃 (π‘₯, 𝑦,π‘š(π‘₯)πœ€, πœ†/πœ€,Β±) analogously toοΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Β±) and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,Β±), only replace οΏ½οΏ½(𝑠) by οΏ½οΏ½(𝑠,π‘šπœ€) in the definition. First we take πœ€ = 1.

22. One-surface reflection. Find 𝑃 (π‘₯, 𝑦,π‘š(π‘₯), πœ†,+) and 𝑃 (π‘₯, 𝑦,π‘š(π‘₯), πœ†,βˆ’) for

π‘š(π‘₯) = π‘š0(π‘₯) ≑ 0 and π‘š(π‘₯) = π‘š1(π‘₯) =

{0.2, for π‘₯ = 0;

0, for π‘₯ = 0.

Now fix odd 𝐿 > 1 called the width of a glass plate. First assume for simplicity that light is reflectedonly by the two surfaces of the plate and thus take11

π‘š2(π‘₯)=

⎧βŽͺ⎨βŽͺβŽ©βˆ’0.2, for π‘₯ = 1;

+0.2, for π‘₯ = 𝐿;

0, otherwise.

The reflection/transmission probabilities12 of light of wavelength πœ† by glass plate of width 𝐿 are respectively𝑃 (πœ†, 𝐿,βˆ’) = lim

𝑦→+βˆžπ‘¦ even

𝑃 (0, 𝑦,π‘š2(π‘₯), πœ†,βˆ’);

𝑃 (πœ†, 𝐿,+) = lim𝑦→+βˆžπ‘¦ even

𝑃 (𝐿 + 1, 𝑦,π‘š2(π‘₯), πœ†,+).

23. Plot the graph of the function 𝑓(𝐿) = 𝑃 (0, 2𝐿,π‘š2(π‘₯), 16,βˆ’).

24.* (Second main problem)Two-surface reflection. Find 𝑃 (πœ†, 𝐿,βˆ’), 𝑃 (πœ†, 𝐿,+), 𝑃 (πœ†, 𝐿,βˆ’)+𝑃 (πœ†, 𝐿,+),and max𝐿 𝑃 (πœ†, 𝐿,βˆ’).

In fact the light is reflected inside the plate; this should be taken into account for a more accuratecomputation of the reflection probability (there is no hope for an exact solution for complicated matter suchas glass). For this purpose we need to modify the model essentially. Take arbitrary πœ€ > 0. Fix π‘š > 0 andlet π‘š3(π‘₯) = π‘š for 0 < π‘₯ ≀ 𝐿/πœ€ and π‘š3(π‘₯) = 0 otherwise. Now for each move starting in a square insidethe glass, our vector is additionally rotated through the angle arctanπ‘šπœ€ clockwise (independently on if thechecker does or does not turn in the square)13. In other words, set οΏ½οΏ½m(𝑠,π‘š(π‘₯)πœ€) := π‘…βˆ’π‘˜ arctanπ‘šπœ€οΏ½οΏ½(𝑠,π‘š(π‘₯)πœ€),where π‘˜ is the number of moves starting in the strip 0 < π‘₯ ≀ 𝐿/πœ€ in the path 𝑠. Define 𝑃m(π‘₯, 𝑦,π‘š(π‘₯)πœ€, πœ†

πœ€,βˆ’)

analogously to 𝑃 (π‘₯, 𝑦,π‘š(π‘₯)πœ€, πœ†πœ€,βˆ’), only replace οΏ½οΏ½(𝑠,π‘š(π‘₯)πœ€) by οΏ½οΏ½m(𝑠,π‘š(π‘₯)πœ€) in the definition.

25.** Thin-film reflection. Find limπœ€β†’0+

lim𝑦→+βˆžπ‘¦ even

𝑃m(0,𝑦,π‘š3(π‘₯)πœ€,πœ†πœ€,βˆ’). For which π‘š maximum of the expres-

sion over 𝐿 equals max𝐿 𝑃 (πœ†, 𝐿,βˆ’)? (Use existence of lim𝑦→+βˆžπ‘¦+π‘₯ even

𝑅2πœ‹π‘¦πœ€/πœ†οΏ½οΏ½(π‘₯,𝑦,π‘š3(π‘₯)πœ€,πœ†πœ€,βˆ’) without proof.)

9Beware: in general Feynman checkerboard is inappropriate to describe light; partial reflection is a remarkable exception.10The mass is proportional to (π‘›βˆ’ 1)/2

βˆšπ‘›, where 𝑛 is the refractive index; for glass 𝑛 β‰ˆ 1.5 and (π‘›βˆ’ 1)/2

βˆšπ‘› β‰ˆ 0.2.

11This simplifying assumption requires negative mass for the left surface; the origin of that becomes clear after solving 25.12It is more conceptual to define 𝑃 (πœ†, 𝐿,βˆ’) = limπœ€β†’0 limΞ”β†’+∞ lim𝑦→+∞

βˆ‘π‘₯∈Z 𝑃 (π‘₯, 𝑦,π‘š3(π‘₯)πœ€,

πœ†πœ€ ,Ξ”,βˆ’) but we do not.

13This additional rotation is explained as follows. The light can be scattered in each square inside the glass several times.Each individual scattering gives a factor of βˆ’π‘–π‘šπœ€ to our vector (viewed as a complex number) and may or may not changethe movement direction. Assume that π‘šπœ€ < 1. Thus a move without changing the direction contributes a factor of

1 (no scattering)βˆ’ π‘–π‘šπœ€ (1 scattering)+ (βˆ’π‘–π‘šπœ€)2 (2 scatterings)+ Β· Β· Β· = 1

1 + π‘–π‘šπœ€.

Without a scattering, the checker moves straightly. Thus a turn in a particular square inside the glass contributes a factor

βˆ’π‘–π‘šπœ€ (1 scattering)+ (βˆ’π‘–π‘šπœ€)2 (2 scatterings)+ (βˆ’π‘–π‘šπœ€)3 (3 scatterings)+ Β· Β· Β· = βˆ’π‘–π‘šπœ€

1 + π‘–π‘šπœ€.

These are the same factors as in the model from Β§3 but additionally rotated through the angle arctanπ‘šπœ€ clockwise.

5

6 External fieldQuestion: what is the probability to find a right electron at (π‘₯, 𝑦), if it moves in a given magnetic field 𝑒?Assumptions: the magnetic field vanishes outside the π‘₯𝑦-plane, it is not affected by the electron.Results: deflection of electron and spin β€˜precession’ in a magnetic field, charge conservation.

Figure 5: Paths in a field

An external magnetic field changes the motion as follows14.A common point of 4 squares of the checkerboard is called a ver-

tex. A magnetic field15 is a fixed assignment 𝑒 of numbers +1 andβˆ’1 to all the vertices. For instance, in Figure 5, the magnetic fieldis βˆ’1 at the top-right vertex of each square (π‘₯, 𝑦) with both π‘₯ and𝑦 even. Modify the definition of the vector οΏ½οΏ½(𝑠) by reversing the di-rection each time when the checker passes through a vertex with themagnetic field βˆ’1. Denote by οΏ½οΏ½(𝑠, 𝑒) the resulting vector. Formally,put οΏ½οΏ½(𝑠, 𝑒) = οΏ½οΏ½(𝑠)𝑒(𝐢1)𝑒(𝐢2) . . . 𝑒(𝐢𝑦), where 𝐢1, 𝐢2, . . . , 𝐢𝑦 are allthe vertices passed by 𝑠. Define οΏ½οΏ½(π‘₯, 𝑦, 𝑒,Β±) and 𝑃 (π‘₯, 𝑦, 𝑒,Β±) analogously to οΏ½οΏ½(π‘₯, 𝑦,Β±) and 𝑃 (π‘₯, 𝑦,Β±) re-placing οΏ½οΏ½(𝑠) by οΏ½οΏ½(𝑠, 𝑒) in the definition. For instance, if 𝑒(𝐢) = +1 identically, then 𝑃 (π‘₯, 𝑦, 𝑒) = 𝑃 (π‘₯, 𝑦).

26. Homogeneous field. Let 𝑒(𝐢) = βˆ’1, if 𝐢 is the top-right vertex of a square (π‘₯, 𝑦) with both π‘₯and 𝑦 even, and 𝑒(𝐢) = +1 otherwise. Find the vector οΏ½οΏ½(π‘₯, 𝑦, 𝑒,+) and the probability 𝑃 (π‘₯, 𝑦, 𝑒,+) for𝑦 = 1, 2, 3, 4 and each integer π‘₯. What is

βˆ‘π‘₯∈Z(𝑃 (π‘₯, 𝑦, 𝑒,+) + 𝑃 (π‘₯, 𝑦, 𝑒,βˆ’)) for fixed 𝑦 = 1, 2, 3, or 4?

27. Spin β€˜precession’ in a magnetic field. Plot the graph of the function 𝑓(𝑦) =βˆ‘

π‘₯∈Z 𝑃 (π‘₯, 𝑦, 𝑒,+)for the field 𝑒 from the previous problem using a computer.

For a given magnetic field 𝑒, a white square is negative, if 𝑒 equals βˆ’1 at 1 or 3 vertices of the square.

28. Gauge transformations. Changing the signs of the values of 𝑒 at the 4 vertices of one black squaresimultaneously does not change 𝑃 (π‘₯, 𝑦, 𝑒,+).

29. Curvature. One can make 𝑒 to be identically +1 in a rectangle formed by checkerboard squares usingthe transformations from Problem 28, if and only if there are no negative white squares in the rectangle.

30. Homology. The magnetic field 𝑒 equals +1 on the boundary of a rectangle π‘š Γ— 𝑛 formed bycheckerboard squares. Which can be the number of negative white squares in the rectangle?

31. Probability/charge conservation. Solve analogues of Problems 4, 5 for 𝑒 not being identically +1.

7 Identical particles and antiparticles

Question: what is the probability to find electrons (or electron+positron) at 𝐹 and 𝐹 β€², emitted from 𝐴 and 𝐴′?Assumptions: the same as in the basic model; 𝑦-coordinate is interpreted as time.Results: exclusion principle.

The motion of several electrons is described by a similar model as follows.To each pair of paths 𝑠, 𝑠′ of the checker, consisting of 𝑦 moves each, assign a vector οΏ½οΏ½(𝑠, 𝑠′) as follows.

Start with the vector (0, 1). Move the checker consecutively along both paths, and rotate the vectoraccording to the same rule as in Β§1: each time when the checker changes the direction, the vector isrotated through 90∘ clockwise. (Thus the vector is rotated totally 𝑑(𝑠) + 𝑑(𝑠′) times, where 𝑑(𝑆) is thenumber of turns in a path 𝑆.) In addition, at the very end the vector is divided by 2π‘¦βˆ’1. The final positionof the vector is denoted by οΏ½οΏ½(𝑠, 𝑠′). For instance, in Figure 2 we have οΏ½οΏ½(𝑠, 𝑠0) = (βˆ’1/4, 0).

Fix squares 𝐴 = (0, 0), 𝐴′ = (π‘₯0, 0), 𝐹 = (π‘₯, 𝑦), 𝐹 β€² = (π‘₯β€², 𝑦) and their diagonal neighbors 𝐡 = (1, 1),𝐡′ = (π‘₯0 + 1, 1), 𝐸 = (π‘₯βˆ’ 1, 𝑦 βˆ’ 1), 𝐸 β€² = (π‘₯β€² βˆ’ 1, 𝑦 βˆ’ 1), where π‘₯0 = 0. Denote16

οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) :=βˆ‘

𝑠:𝐴𝐡→𝐸𝐹𝑠′:𝐴′𝐡′→𝐸′𝐹 β€²

οΏ½οΏ½(𝑠, 𝑠′) βˆ’βˆ‘

𝑠:𝐴𝐡→𝐸′𝐹 ′𝑠′:𝐴′𝐡′→𝐸𝐹

οΏ½οΏ½(𝑠, 𝑠′),

14Beware that this method of adding the magnetic field, although well-known, is very different from the one from [Feynman].15Or electromagnetic vector-potential, to be precise. The field is interpreted as magnetic or electric depending on if the

𝑦-coordinate is interpreted as position or time.16Here it is essential that 𝑠 and 𝑠′ are paths of particles of the same sort, e.g., two electrons. Otherwise the 2nd sum is

omitted. The sign before the 2nd sum is changed to plus for some other sorts of particles, e.g., photons (particles of light).

6

where the first sum is over all pairs consisting of a checker path 𝑠 starting with the move 𝐴𝐡 and endingwith the move 𝐸𝐹 , and a path 𝑠′ starting with the move 𝐴′𝐡′ and ending with the move 𝐸 ′𝐹 β€², whereasin the second sum the final moves are interchanged.

The length square 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) = |π‘Ž(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²)|2 is called the probability17

to find right electrons at 𝐹 and 𝐹 β€², if they are emitted from 𝐴 and 𝐴′. In particular, 𝑃 (𝐴𝐡,𝐴′𝐡′ →𝐸𝐹,𝐸𝐹 ) = 0, i.e., two right electrons cannot be found at the same point; this is called exclusion principle.

32. Independence. For π‘₯0 β‰₯ 2𝑦 and π‘₯β€² > π‘₯ express 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) through 𝑃 (π‘₯, 𝑦,+) and𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,+).

33. Exclusion principle (for intermediate states). Prove that οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) is notchanged, if the sums in the definition are over only those pairs of paths 𝑠, 𝑠′ which have no common moves.

Define 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) analogously also for 𝐸 = (π‘₯Β± 1, 𝑦 βˆ’ 1), 𝐸 β€² = (π‘₯β€² Β± 1, 𝑦 βˆ’ 1).

34. Probability/charge conservation. We haveβˆ‘

𝐸,𝐸′,𝐹,𝐹 β€² 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) = 1, where thesum is over all quadruples 𝐹 = (π‘₯, 𝑦), 𝐹 β€² = (π‘₯β€², 𝑦), 𝐸 = (π‘₯Β±1, π‘¦βˆ’1), 𝐸 β€² = (π‘₯β€²Β±1, π‘¦βˆ’1) with fixed 𝑦 β‰₯ 1.

An electron has an antiparticle called positron. One can think of an antiparticle as a particle movingbackwards in time. The motion is described by a similar model as follows.

Add one more checker to the checkerboard, which now moves either downwards-right or downwards-leftto the diagonal neighbors. The added checker is called black, while the one studied before is called white.

For each pair of paths 𝑠, 𝑠′ of the white and black checker respectively, define a vector οΏ½οΏ½(𝑠, 𝑠′) analogouslyto the above, only for each turn of the black checker rotate the vector counterclockwise rather than clockwise(independently of the direction the checker turns). For instance, οΏ½οΏ½(𝑠, 𝑠′) = (0, 21βˆ’π‘¦), if 𝑠 and 𝑠′ is the samepath with opposite directions. Denote18 by

οΏ½οΏ½(𝐴𝐡,𝐡′𝐴′ β†’ 𝐸𝐹, 𝐹 ′𝐸 β€²) :=βˆ‘π‘ ,𝑠′

οΏ½οΏ½(𝑠, 𝑠′)

the sum is over all the pairs of a white-checker path 𝑠 starting with the move 𝐴𝐡 and ending with themove 𝐸𝐹 , and a black-checker path 𝑠′ starting with the move 𝐹 ′𝐸 β€² and ending with the move 𝐡′𝐴′. Thelength square 𝑃 (𝐴𝐡,𝐡′𝐴′ β†’ 𝐸𝐹, 𝐹 ′𝐸 β€²) = |π‘Ž(𝐴𝐡,𝐡′𝐴′ β†’ 𝐸𝐹, 𝐹 ′𝐸 β€²)|2 is called the probability to find anelectron at 𝐹 and a positron at 𝐹 β€², if they are emitted from 𝐴 and 𝐴′.

35. Independence. Solve analogue of Problem 32 for 𝑃 (𝐴𝐡,𝐡′𝐴′ β†’ 𝐸𝐹, 𝐹 ′𝐸 β€²).

8 Interaction***Question: the same as in the basic model but the 𝑦-coordinate is now interpreted as time.Assumptions: the electron now generates a magnetic field affecting the motion; encircling by waveproof walls.Results: interaction with the walls changes the motion of an electron.

We have reached an unexplored area: a physically correct rigorous definition of the next upgrade isnot yet known. In this section we give a WRONG definition leading to paradoxical results:

βˆ™ the upgrades measures the interaction of the electron with the walls rather than self-interaction;

βˆ™ the interaction (with the walls) propagates at infinite speed rather than the speed of light.

But we hope that the introduced ideas still might be of interest.The moving electron itself generates an magnetic field, which in turn affects the motion. The generated

field is random, with the probability resulting from summation over all possible intermediate fields.Fix 𝑔 β‰₯ 0 called the interaction constant19 (it is related to the electron charge). Fix a black square

(π‘₯, 𝑦) called the final position. Fix the rectangle 𝑅 formed by all the squares (π‘₯β€², 𝑦′) such that 1βˆ’π‘¦ < π‘₯β€² < 𝑦

17One should think of the value 𝑦 as fixed, and the quadruples (𝐹, 𝐹 β€², 𝐸,𝐸′) as the possible outcomes of an experiment.18This definition makes sense only for π‘₯0 β‰₯ 2𝑦; otherwise annihilation of particles cannot be ignored, and the model

becomes inappropruate.19If the lattice step πœ– is not fixed, then 𝑔 might depend on it. This is called renormalization; we do not discuss it.

7

and 0 < 𝑦′ < 𝑦. This waveproof box 𝑅 encircles any possible checker path 𝑠 from (0, 0) to (π‘₯, 𝑦) excludingthe first and the last move; it is required to make the sum over all intermediate fields below finite.

Take any such path 𝑠 and any assignment 𝑒 of the numbers Β±1 to all the vertices in the rectangle 𝑅.Let 𝑛 be the number of negative white squares in 𝑅 for 𝑒 (they play the role of turnings of the checker).Denote by

οΏ½οΏ½(𝑠, 𝑒, 𝑔) =𝑔𝑛

2(π‘¦βˆ’2)2(1 + 𝑔2)(π‘¦βˆ’1)2/2π‘…βˆ’πœ‹π‘›/2οΏ½οΏ½(𝑠, 𝑒)

the vector20 οΏ½οΏ½(𝑠, 𝑒), rotated clockwise through 𝑛·90∘, multiplied by 𝑔𝑛 and divided by 2(π‘¦βˆ’2)2(1+𝑔2)(π‘¦βˆ’1)2/2.Let 𝑒fin be any assignment of the numbers Β±1 to all the vertices on the top side of 𝑅 such that 𝑒fin

equals +1 at the endpoints of the side. It is called the final magnetic field. Denote by

οΏ½οΏ½(π‘₯, 𝑦, 𝑒fin, 𝑔,+) :=βˆ‘π‘ ,𝑒

οΏ½οΏ½(𝑠, 𝑒, 𝑔)

the sum over all the paths 𝑠 from (0, 0) to (π‘₯, 𝑦) starting and ending with an upwards-right move and overall the assignments 𝑒 of the numbers Β±1 to all the vertices in the rectangle 𝑅 such that

𝑒 =

{𝑒fin on the top side of 𝑅;

+1 on all the other sides of 𝑅.

The length square 𝑃 (π‘₯, 𝑦, 𝑒fin, 𝑔,+) = |π‘Ž(π‘₯, 𝑦, 𝑒fin, 𝑔,+)|2 is the probability that the final magnetic fieldequals 𝑒fin and a right electron is found at (π‘₯, 𝑦). Define 𝑃 (π‘₯, 𝑦, 𝑒fin, 𝑔,βˆ’) analogously.

36. Find the probabilities 𝑃 (π‘₯, 𝑦, 𝑒fin, 1,+) and 𝑃 (π‘₯, 𝑦, 𝑒fin, 1,βˆ’) for 𝑔=1, 𝑦=2, 3 and all possible π‘₯ and 𝑒fin.

The sum21

𝑃 (π‘₯, 𝑦, 𝑔,+) =βˆ‘π‘’fin

𝑃 (π‘₯, 𝑦, 𝑒fin, 𝑔,+)

over all assignments 𝑒fin of the numbers Β±1 to the vertices on the top side of 𝑅 is the probability to find aright electron at (π‘₯, 𝑦). Define 𝑃 (π‘₯, 𝑦, 𝑔,βˆ’) analogously. For instance, 𝑃 (π‘₯, 𝑦, 0,+) = 𝑃 (π‘₯, 𝑦,+) (why?).

37.* For a path 𝑠 from (0, 0) to (π‘₯, 𝑦), consider the sumβˆ‘

𝑒 οΏ½οΏ½(𝑠, 𝑒, 𝑔) over all assignments 𝑒 of Β±1 to allthe vertices in the rectangle 𝑅 such that 𝑒 = +1 on the boundary of 𝑅. Express the sum through οΏ½οΏ½(𝑠)and the number of white squares in each of the two parts, into which 𝑅 is divided by the path 𝑠.

38.* Let 𝑒± be equal to Β±1 respectively at the top-right corner of the square (2 βˆ’ 𝑦, 𝑦 βˆ’ 1), and equal to+1 at all the other vertices on the top side of 𝑅. Solve analogues of Problems 4 and 5 for οΏ½οΏ½(π‘₯, 𝑦, 𝑒±, 𝑔,Β±)and 𝑃 (π‘₯, 𝑦, 𝑔,+) + 𝑃 (π‘₯, 𝑦, 𝑔,βˆ’) instead of π‘Ž1(π‘₯, 𝑦), π‘Ž2(π‘₯, 𝑦), and 𝑃 (π‘₯, 𝑦).

9 Creation and annihilation***Question: the same as in the model with identical particles and antiparticles.Assumptions: electron-positron pairs now created and annihilated, no interaction, encircling by reflecting walls.Results: no; this is only an ingredient for more realistic models with interaction.

Finally we have reached an unexplored area: we state an almost 40-years-old open problem.Start with mentioning what is not done in this section:

βˆ™ we do not give a definition of the new upgrade (it is unknown so far);

βˆ™ the upgrade (even if defined) would not explain any new experimental results.

But

βˆ™ we do give a precise statement of the problem: which exactly properties of the upgrade are requested;

20It is well-defined because the path 𝑠 is contained in 𝑅 except the first and the last move. We set 𝑔𝑛 = 1 for 𝑔 = 𝑛 = 0.21Here we sum probabilities rather than vectors. The notation 𝑃 (π‘₯, 𝑦, 𝑔,+) should not be confused with 𝑃 (π‘₯, 𝑦,π‘šπœ€,+).

8

βˆ™ the upgrade is an important ingredient of further ones fantastically agreeing with experiment.

Informally, our plan is as follows. Checker paths turning downwards or upwards or forming cycles meancreation and annihilation of electron-positron pairs. Even if we start with just one electron, we might endup with many electrons and positrons. To each possible configuration of the resulting particles, we wantto assign a complex number so that its length square is the probability of the configuration in a sense. Thenumber itself is the sum over all possible transitions from the initial configuration to the final one, that is,all possible paths configurations joining them. To make the sum finite, we put reflecting walls around.

Fix a rectangle 𝑅 formed by all the squares (π‘₯, 𝑦) such that π‘₯min ≀ π‘₯ ≀ π‘₯max and 0 ≀ 𝑦 ≀ 𝑦max (thelines π‘₯ = π‘₯min and π‘₯ = π‘₯max are called reflecting walls). Fix π‘š, πœ€ > 0 called mass and lattice step.

An initial configuration is any assignment22 of one of the signs β€œ+” or β€œβˆ’β€ to some vertices inside 𝑅lying between the lines 𝑦 = 0 and 𝑦 = 1. Physically the signs mean the initial positions of positrons andelectrons respectively (and points without a sign are vacant). For our game, this means that the checkerspass the vertices with the β€œβˆ’β€ sign upwards-left or -right, and the vertices with the β€œ+” sign β€” downwards-left or -right (and do not pass through vertices without a sign). Analogously, a final configuration is anassignment to vertices between the lines 𝑦 = 𝑦max and 𝑦 = 𝑦max βˆ’ 1. An intermediate configuration is anyassignment of one of the signs β€œ+” or β€œβˆ’β€ to some vertices inside 𝑅 such that the difference between thenumber of β€œ+” and β€œβˆ’β€ signs on the 2 top vertices of each black square in the strip 1 ≀ 𝑦 ≀ 𝑦maxβˆ’1 equalsthe difference on the 2 bottom vertices. For our game, this means that the checkers start and finish motionin the lines 𝑦 = 0 or 𝑦 = 𝑦max only. In other words, the signs at the vertices of each black square are inone of the 19 positions23 shown in Figure 6 to the left. These 19 positions are called basic configurations.

Figure 6: Basic configurations

Suppose that one has assigned a complex number (depending on π‘šπœ€) to each of the 19 basic configu-rations. A β€œright” choice of the numbers is unknown; think of them as fixed parameters of our model.

Then take any intermediate configuration. In each black square inside 𝑅 not having common pointswith the boundary, write the complex number assigned to the basic configuration in the black square.Assign the product of all the written numbers to the intermediate configuration.

Now to any pair of initial and final configurations, sum the complex numbers assigned to all possibleintermediate configurations between them. Assign the resulting complex number to the pair.

39. (Puzzle) Restrict to configurations without β€œ+” signs. Let 𝐴,𝐴′, 𝐡,𝐡′, 𝐸, 𝐸 β€², 𝐹, 𝐹 β€² be the black squaresdefined in Β§7. Take any π‘₯min < βˆ’π‘¦ and π‘₯max > π‘₯0 + 𝑦. Fix the initial and final configurations with exactlytwo β€œβˆ’β€ signs, located at the top-right vertices of the squares 𝐴, 𝐴′, and 𝐸, 𝐸 β€² respectively. Assign complexnumbers to the 6 basic configurations without β€œ+” signs so that the sum of the numbers assigned to allintermediate configurations without β€œ+” signs equals to the vector οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸 ′𝐹 β€²) from Β§7.

A case of particular interest is when the initial configuration consists of just one β€œβˆ’β€ sign at the top-right vertex of the square (0, 0) and the final configuration consists of just one β€œβˆ’β€ sign at the bottom-right

22The assignment has nothing to do with the magnetic field from Β§6.23We consider positions of signs, but not possible checker paths in a particular black square like in Figure 6 to the right.

9

or bottom-left vertex of a black square (π‘₯, 𝑦max). The complex numbers assigned to the pairs in questionare denoted by οΏ½οΏ½(π‘₯, 𝑦max,π‘šπœ€, π‘₯min, π‘₯max,βˆ’) and οΏ½οΏ½(π‘₯, 𝑦max,π‘šπœ€, π‘₯min, π‘₯max,+) respectively.

The desired continuum limit of these complex numbers involves the following modified Bessel functionsand Hankel functions :

𝐾0(π‘₯) =

∫ ∞

1

π‘’βˆ’π‘₯𝑑

βˆšπ‘‘2 βˆ’ 1

𝑑𝑑 𝐻(1)0 (π‘₯) =

2

π‘–πœ‹

∫ ∞

1

𝑒𝑖π‘₯π‘‘βˆšπ‘‘2 βˆ’ 1

𝑑𝑑

𝐾1(π‘₯) = π‘₯

∫ ∞

1

π‘’βˆ’π‘₯π‘‘βˆšπ‘‘2 βˆ’ 1 𝑑𝑑 𝐻

(1)1 (π‘₯) = βˆ’2π‘₯

π‘–πœ‹

∫ ∞

1

𝑒𝑖π‘₯π‘‘βˆšπ‘‘2 βˆ’ 1 𝑑𝑑

40.*** Continuum limit. Assign complex numbers to the 19 basic configurations so that for each |𝑦| < |π‘₯|

limπ‘›β†’βˆž

limπ‘₯maxβ†’βˆž

𝑛 οΏ½οΏ½(

2βŒŠπ‘›π‘₯

2

βŒ‹, 2βŒŠπ‘›π‘¦

2

βŒ‹,π‘š

𝑛,βˆ’π‘₯max, π‘₯max,βˆ’

)= π‘šπΎ0(π‘š

√π‘₯2 βˆ’ 𝑦2);

limπ‘›β†’βˆž

limπ‘₯maxβ†’βˆž

𝑛 οΏ½οΏ½(

2βŒŠπ‘›π‘₯

2

βŒ‹, 2βŒŠπ‘›π‘¦

2

βŒ‹,π‘š

𝑛,βˆ’π‘₯max, π‘₯max,+

)= βˆ’π‘–π‘š

π‘₯ + π‘¦βˆšπ‘₯2 βˆ’ 𝑦2

𝐾1(π‘šβˆšπ‘₯2 βˆ’ 𝑦2);

and for each |𝑦| > |π‘₯| we have

limπ‘›β†’βˆž

limπ‘₯maxβ†’βˆž

𝑛 οΏ½οΏ½(

2βŒŠπ‘›π‘₯

2

βŒ‹, 2βŒŠπ‘›π‘¦

2

βŒ‹,π‘š

𝑛,βˆ’π‘₯max, π‘₯max,βˆ’

)= π‘–π‘š

πœ‹

2𝐻

(1)0 (π‘š

βˆšπ‘¦2 βˆ’ π‘₯2);

limπ‘›β†’βˆž

limπ‘₯maxβ†’βˆž

𝑛 οΏ½οΏ½(

2βŒŠπ‘›π‘₯

2

βŒ‹, 2βŒŠπ‘›π‘¦

2

βŒ‹,π‘š

𝑛,βˆ’π‘₯max, π‘₯max,+

)= π‘š

πœ‹

2Β· π‘₯ + π‘¦βˆš

𝑦2 βˆ’ π‘₯2𝐻

(1)1 (π‘š

βˆšπ‘¦2 βˆ’ π‘₯2).

Let us discuss the physical meaning of the upgrade. For 𝑦 ≫ |π‘₯| the right-hand sides of the lattertwo equations are very close to the right-hand sides of the answer to Problem 10. Thus one may wish tointerpret the upgrade as a more accurate approximation for the probability to find the electron in a square(π‘₯, 𝑦). But this faces serious objections.

First, it is in principle impossible to measure the coordinates of an electron exactly 24. Such a prioriuncertainty has the same order of magnitude as the correction introduced by the upgrade. Thus theupgrade does not actually add anything to description of the electron motion.

Second, for fixed initial configuration, the squares of the absolute values of the numbers assigned toall the possible final configurations do not sum up to 1 (even in the continuum limit). The reason is thatdistinct configurations are not mutually exclusive: for a field in a given configuration, there is a positiveprobability to find it in a different configuration. We do not know any clear explanation of that.

To summarize, the upgrade lacks a direct physical interpretation, and should be considered as aningredient for further upgrades.

10 (1 + 1)-dimensional quantum electrodynamics***

Question: what is the probability to find electrons (or electron+positron) with momenta π‘ž and π‘žβ€² in the far future,if they were emitted with momenta 𝑝 and 𝑝′ in the far past?Assumptions: interaction now switched on; all simplifying assumptions removed except the default ones:no nuclear forces, no gravitation, electron moves only along the π‘₯-axis (and 𝑦-coordinate is interpreted as time).Results: quantum corrections.

Unifying the (so far unknown) upgrades discussed in the previous two sections would give an elementarydefinition of (1 + 1)-dimensional QED.

Future research

An algorithmic quantum field theory is a one which for each experimentally observable quantity and apositive number πœ€ provides a precise statement of an algorithm giving the predicted value of the quantitywithin accuracy πœ€. (Surely, the predicted value does not have to agree with the experiment for πœ€ less thanaccuracy of theory itself.) This is an extension of constructive quantum field theory, the latter currentlybeing far away from algorithmic one.

24This should not be confused with uncertainty principle, which does not allow simultaneous measurement of the coordi-nates and momentum.

10

Epilogue (underwater rocks)

We hope that at least some of our readers have become interested in elementary particles and want tolearn more about them. As an epilogue, let us give a few warnings to such readers.

In popular science, theory of elementary particles is usually oversimplified. This sequence of problemsis not an exception. The toy models introduced here are very rough and should be considered with agrain of salt. Simplicity is their only advantage; if taken too seriously, the models could even give a wrongphysical intuition. Real understanding of particles theory requires excellent knowledge of both physics andmathematics.

We should also remark that nowadays there are almost no mathematical results in lattice quantumfield theory; what we have is usually just a numeric simulation. Finally, there are β€œtheories of NewPhysics” which are developed without any objective truth criterion: such theories are supported by neitherexperimental nor mathematical proofs (and some of them have experimental disproofs).

11

Hints, solutions, answers

For any vector οΏ½οΏ½ ∈ R2 denote by π‘Ž1, π‘Ž2 the coordinates of οΏ½οΏ½. E.g., οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€,βˆ’) = (π‘Ž1(π‘₯, 𝑦,π‘šπœ€,βˆ’), π‘Ž2(π‘₯, 𝑦,π‘šπœ€,βˆ’)).Sometimes the vector οΏ½οΏ½ is considered as a complex number π‘Ž1 + π‘–π‘Ž2 (although complex numbers are not requiredfor the solution of most problems). In what follows assume that π‘₯ and 𝑦 have the same parity unless the oppositeis indicated.

1. Answer. In the table, the vector in the cell (π‘₯, 𝑦) is οΏ½οΏ½(π‘₯, 𝑦), and an empty cell (π‘₯, 𝑦) means that οΏ½οΏ½(π‘₯, 𝑦) = (0, 0).

4(

12√2, 0) (

0,βˆ’ 12√2

) (1

2√2,βˆ’ 1√

2

) (0, 1

2√2

)3

(12 , 0

) (12 ,βˆ’

12

) (0, 12

)2

(1√2, 0) (

0, 1√2

)1 (0, 1)

y x βˆ’2 βˆ’1 0 1 2 3 4

In the following table, the number in the cell (π‘₯, 𝑦) is 𝑃 (π‘₯, 𝑦), and an empty cell (π‘₯, 𝑦) means that 𝑃 (π‘₯, 𝑦) = 0.

4 1/8 1/8 5/8 1/8

3 1/4 1/2 1/4

2 1/2 1/2

1 1

y x βˆ’2 βˆ’1 0 1 2 3 4

(The vectors οΏ½οΏ½(π‘₯, 𝑦) can be easily computed consecutively using Problem 4.) See also the following figure.

For any positive 𝑦 we haveβˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦) = 1. (This is Problem 5.)

For βˆ’4 ≀ βˆ’π‘¦ < π‘₯ ≀ 𝑦 ≀ 4 the probability 𝑃 (π‘₯, 𝑦) equals 0 if and only if one of the numbers π‘₯, 𝑦 is odd and theother one is even. For any π‘₯ > 𝑦 and π‘₯ < βˆ’π‘¦ + 1 we have 𝑃 (π‘₯, 𝑦) = 0. (In general it is not known, if 𝑃 (π‘₯, 𝑦) = 0can vanish for π‘₯ and 𝑦 of the same parity satisfying βˆ’π‘¦ < π‘₯ ≀ 𝑦.)

For each odd 𝑦 we have οΏ½οΏ½(0, 𝑦) = 0. For each nonnegative integer 𝑛 we have

the vector οΏ½οΏ½(0, 8𝑛+ 2) is directed to the right,

the vector οΏ½οΏ½(0, 8𝑛+ 4) is directed downwards,

the vector οΏ½οΏ½(0, 8𝑛+ 6) is directed to the left,

the vector οΏ½οΏ½(0, 8𝑛+ 8) is directed upwards.

Analogously, for each even 𝑦 we have οΏ½οΏ½(1, 𝑦) = 0. For each nonnegative integer 𝑛 we have

the vector οΏ½οΏ½(1, 8𝑛+ 3) is directed downwards-right,

the vector οΏ½οΏ½(1, 8𝑛+ 5) is directed downwards-left,

the vector οΏ½οΏ½(1, 8𝑛+ 7) is directed upwards-left,

the vector οΏ½οΏ½(1, 8𝑛+ 9) is directed upwards-right.

(See the remark after the answer to Problem 9.)

12

Remark. In the table, 2(𝑛+π‘˜βˆ’2)/2οΏ½οΏ½(π‘˜βˆ’π‘›+1, 𝑛+ π‘˜βˆ’ 1) stands at the intersection of π‘˜-th column and 𝑛-th row:

(0, 1) (0, 1) (0, 1) (0, 1) (0, 1) (0, 1) (0, 1) (0, 1)(1, 0) (1,βˆ’1) (1,βˆ’2) (1,βˆ’3) (1,βˆ’4) (1,βˆ’5) (1,βˆ’6) (1,βˆ’7)(1, 0) (0,βˆ’1) (βˆ’1,βˆ’1) (βˆ’2, 0) (βˆ’3, 2) (βˆ’4, 5) (βˆ’5, 9) (βˆ’6, 14)(1, 0) (βˆ’1,βˆ’1) (βˆ’2, 0) (βˆ’2, 2) (βˆ’1, 4) (1, 5) (4, 4) (8, 0)(1, 0) (βˆ’2,βˆ’1) (βˆ’2, 1) (0, 3) (3, 3) (6, 0) (8,βˆ’6) (8,βˆ’14)(1, 0) (βˆ’3,βˆ’1) (βˆ’1, 2) (3, 3) (6, 0) (6,βˆ’6) (2,βˆ’12) (βˆ’6,βˆ’14)(1, 0) (βˆ’4,βˆ’1) (1, 3) (6, 2) (6,βˆ’4) (0,βˆ’10) (βˆ’10,βˆ’10) (βˆ’20, 0)(1, 0) (βˆ’5,βˆ’1) (4, 4) (8, 0) (2,βˆ’8) (βˆ’10,βˆ’10) (βˆ’20, 0) (βˆ’20, 20)

2. Answer: no for both questions. For example, 𝑃 (2, 4) = 5/8 = 1/8+1/4 = 𝑃 (2, 4 bypass 2, 2)+𝑃 (2, 4 bypass 0, 2)and 𝑃 (0, 4) = 1/8 < 1/4 = 𝑃 (0, 4 bypass 2, 2).

3. Answer: 𝑃 (0, 12) = 25/512. Hint: the answer is obtained immediately by means of Remark after the answerto Problem 9 or it can be quickly computed recursively by means of the answer to Problem 4.

4. Answer:

{π‘Ž1(π‘₯, 𝑦) = 1√

2π‘Ž2(π‘₯+ 1, 𝑦 βˆ’ 1) + 1√

2π‘Ž1(π‘₯+ 1, 𝑦 βˆ’ 1);

π‘Ž2(π‘₯, 𝑦) = 1√2π‘Ž2(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ 1√

2π‘Ž1(π‘₯βˆ’ 1, 𝑦 βˆ’ 1).

Solution. Let us derive the formula for π‘Ž2(π‘₯, 𝑦); the derivation for π‘Ž1(π‘₯, 𝑦) is analogous. Consider anypath 𝑠 from (0, 0) to (π‘₯, 𝑦). Denote by 𝑑(𝑠) the number of turnings in 𝑠. Denote οΏ½οΏ½(𝑠) = (π‘Ž1(𝑠), π‘Ž2(𝑠)). Noticethat the component π‘Ž2(𝑠) = 0 if and only if 𝑑(𝑠) is even and π‘Ž1(𝑠) = 0 if and only if 𝑑(𝑠) is odd. Therefore,π‘Ž2(π‘₯, 𝑦) =

βˆ‘π‘ :𝑑(𝑠) even

π‘Ž2(𝑠) and π‘Ž1(π‘₯, 𝑦) =βˆ‘

𝑠:𝑑(𝑠) odd

π‘Ž1(𝑠).

The last move in the path 𝑠 is made either from (π‘₯βˆ’ 1, 𝑦) or from (π‘₯+ 1, 𝑦). It is obvious that if 𝑑(𝑠) is even,then the last move is directed upwards-right, else it is directed upwards-left. Since we are interested in π‘Ž2(π‘₯, 𝑦),assume that last move is directed upwards-right.

Denote by 𝑠′ the path 𝑠 without the last move. If the directions of the last moves in 𝑠 and 𝑠′ coincide, thenοΏ½οΏ½(𝑠) = 1√

2οΏ½οΏ½(𝑠′), otherwise οΏ½οΏ½(𝑠) = 1√

2(π‘Ž2(𝑠

β€²),βˆ’π‘Ž1(𝑠′)). Therefore,

π‘Ž2(π‘₯, 𝑦) =βˆ‘

𝑠:𝑑(𝑠) even

π‘Ž2(𝑠) =1√2

βŽ›βŽ βˆ‘π‘ β€²:𝑑(𝑠′) even

π‘Ž2(𝑠′)βˆ’

βˆ‘π‘ β€²:𝑑(𝑠′) odd

π‘Ž1(𝑠′)

⎞⎠ =1√2(π‘Ž2(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ π‘Ž1(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)).

5. Let us prove by induction over 𝑦 thatβˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦) = 1 for all 𝑦 β‰₯ 1: Obviously,βˆ‘π‘₯∈Z

𝑃 (π‘₯, 1) = 1. The step of

induction follows immediately from the following computation:βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦 + 1) =βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦 + 1)2 + π‘Ž2(π‘₯, 𝑦 + 1)2

]=

βˆ‘π‘₯∈Z

π‘Ž1(π‘₯, 𝑦 + 1)2 +βˆ‘π‘₯∈Z

π‘Ž2(π‘₯, 𝑦 + 1)2 =

=1

2

βˆ‘π‘₯∈Z

(π‘Ž1(π‘₯+ 1, 𝑦) + π‘Ž2(π‘₯+ 1, 𝑦))2 +1

2

βˆ‘π‘₯∈Z

(π‘Ž2(π‘₯βˆ’ 1, 𝑦)βˆ’ π‘Ž1(π‘₯βˆ’ 1, 𝑦))2 =

=1

2

βˆ‘π‘₯∈Z

(π‘Ž1(π‘₯, 𝑦) + π‘Ž2(π‘₯, 𝑦))2 +

1

2

βˆ‘π‘₯∈Z

(π‘Ž2(π‘₯, 𝑦)βˆ’ π‘Ž1(π‘₯, 𝑦))2 =

βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦)

2 + π‘Ž2(π‘₯, 𝑦)2]=

βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦).

A generalization of conservation law by Gleb Minaev and Ivan Russkikh, participants of Summer conference of

Tournament of towns. Perform the change coordinates (π‘₯, 𝑦) ↦→ (𝑑, 𝑒) = (π‘₯+𝑦2 βˆ’ 1, π‘¦βˆ’π‘₯

2 ), i.e., rotate the coordinatesystem through 45 degrees clockwise and shift it by the vector (βˆ’1, 0).

Denote οΏ½οΏ½(𝑑, 𝑒) := οΏ½οΏ½(π‘‘βˆ’ 𝑒+ 1, 𝑑+ 𝑒+ 1), 𝑄(𝑑, 𝑒) := 𝑃 (π‘‘βˆ’ 𝑒+ 1, 𝑑+ 𝑒+ 1), οΏ½οΏ½(𝑑, 𝑒) := (1 +π‘š2πœ€2)(𝑑+𝑒)/2 οΏ½οΏ½(𝑑, 𝑒).The coordinates of the vectors οΏ½οΏ½(𝑑, 𝑒) and οΏ½οΏ½(𝑑, 𝑒) denote by 𝑏1(𝑑, 𝑒), 𝑏2(𝑑, 𝑒), 𝐡1(𝑑, 𝑒), 𝐡2(𝑑, 𝑒).

Remark. In the new coordinate system, a checker moves between neighbouring points of the integer latticerather than along the diagonals between black squares. Also we suppose that we start at (0, 0) and move to anypoint of (Nβˆͺ{0})2, and the additional β€œpre-move” from (βˆ’1, 0) to (0, 0) is taken into account only to compute thenumber of turns.

For a subset 𝑀 βŠ‚ (N βˆͺ {0})2 denote οΏ½οΏ½(𝑑, 𝑒 bypass 𝑀) :=βˆ‘

𝑠 οΏ½οΏ½(𝑠), where we sum is over all paths 𝑠 from

(0, 0) to (𝑑, 𝑒), which bypass the points of the set 𝑀 . Analogously, define 𝑄(𝑑, 𝑒 bypass 𝑀), οΏ½οΏ½(𝑑, 𝑒 bypass 𝑀,Β±),𝑄(𝑑, 𝑒 bypass 𝑀,Β±). Also for a set 𝑀 denote

𝑄(𝑀) :=βˆ‘π‘βˆˆπ‘€

𝑄(𝑝 bypass 𝑀 βˆ– {𝑝}).

13

Remark. Note that for an infinite set 𝑀 the sum becomes infinite as well. The order of a summation isirrelevant because all the summands are positive.

Theorem 1. For each finite set 𝑀 βŠ‚ (N βˆͺ {0})2 such that there are no infinite paths from (0; 0) bypassing the

points of 𝑀 we have 𝑄(𝑀) = 1.

Proof. Prove the theorem by induction overπ‘š := max(𝑑,𝑒)βˆˆπ‘€ (𝑑+𝑒), i.e., the maximal number of a downwards-rightdiagonal containing at least one point of the set 𝑀 . The diagonal is called maximal.

Base: π‘š = 0. In this case 𝑀 = (0, 0) and 𝑄(𝑀) = 𝑄(0, 0) = 1.Step. Let us prove a lemma.

Lemma 1. If a set 𝐴 βŠ‚ (N∩{0})2, which does not contain the points (𝑑, 𝑒), (𝑑, 𝑒+1), (𝑑+1, 𝑒) ∈ (N∩{0})2 , then

𝑄(𝑑, 𝑒 bypass 𝐴) = 𝑄(𝑑, 𝑒+ 1 bypass 𝐴,βˆ’) +𝑄(𝑑+ 1, 𝑒 bypass 𝐴,+).

Proof. This is straightforward:

𝑄(𝑑, 𝑒+ 1 bypass 𝐴,βˆ’) +𝑄(𝑑+ 1, 𝑒 bypass 𝐴,+) = 𝑏1(𝑑, 𝑒+ 1 bypass 𝐴)2 + 𝑏2(𝑑+ 1, 𝑒 bypass 𝐴)2 =

=(𝑏1(𝑑, 𝑒 bypass 𝐴) + 𝑏2(𝑑, 𝑒 bypass 𝐴))2 + (𝑏2(𝑑, 𝑒 bypass 𝐴)βˆ’ 𝑏1(𝑑, 𝑒 bypass 𝐴))2

2=

= 𝑏1(𝑑, 𝑒 bypass 𝐴)2 + 𝑏2(𝑑, 𝑒 bypass 𝐴)2 = 𝑄(𝑑, 𝑒 bypass 𝐴).

Suppose that the point (𝑑, 𝑒) ∈ 𝑀 is such that 𝑑+ 𝑒 is maximal. Suppose that there is a checker path startingat the point (0, 0) and ending at (𝑑, 𝑒) with the last move, say, in the upwards direction bypassing all the pointsof the set 𝑀 βˆ– {(𝑑, 𝑒)}. Then there exists a path to the point (𝑑, 𝑒 βˆ’ 1), bypassing all the points of the set𝑀 βˆ– {(𝑑, 𝑒 βˆ’ 1)}. Notice that in this case (𝑑, 𝑒 βˆ’ 1) /∈ 𝑀 , because the move to (𝑑;𝑒) is in the upwards direction.Hence there exists a path to the point (𝑑 + 1, 𝑒 βˆ’ 1), bypassing all the points of the set 𝑀 βˆ– {(𝑑 + 1, 𝑒 βˆ’ 1)}. If(𝑑 + 1, 𝑒 βˆ’ 1) /∈ 𝑀 then there exists a path going to infinity passing through (𝑑 + 1, 𝑒 βˆ’ 1) and bypassing all thepoints of the set 𝑀 . For example, the path turning right at the point (𝑑, 𝑒) and only going right from there passesthrough (𝑑, 𝑒) and bypasses all the points in 𝑀 because the diagonal is maximal. Therefore (𝑑 + 1, 𝑒 βˆ’ 1) ∈ 𝑀 .Notice that 𝑄(π‘Ž, 𝑏 bypass 𝐴) = 𝑄(π‘Ž, 𝑏 bypass 𝐴,+) + 𝑄(π‘Ž, 𝑏 bypass 𝐴,βˆ’) for any set 𝐴 βŠ‚ (N ∩ {0})2. Denote𝐾 := 𝑀 βˆ–{(𝑑, 𝑒); (𝑑+1, π‘’βˆ’1)}. Then we have the following chain of equalities (where we use 1 for the set 𝐴 = 𝐾):

𝑄(𝑀) = 𝑄(𝐾) +𝑄(𝑑, 𝑒 bypass 𝑀 βˆ– {(𝑑, 𝑒)}) +𝑄(𝑑+ 1, π‘’βˆ’ 1 bypass 𝑀 βˆ– {(𝑑+ 1, π‘’βˆ’ 1)}) == 𝑄(𝐾) +𝑄(𝑑, 𝑒 bypass 𝑀 βˆ– {(𝑑, 𝑒)},+) +𝑄(𝑑, 𝑒 bypass 𝑀 βˆ– {(𝑑, 𝑒)},βˆ’)+

+𝑄(𝑑+ 1, π‘’βˆ’ 1 bypass 𝑀 βˆ– {(𝑑+ 1, π‘’βˆ’ 1)},+) +𝑄(𝑑+ 1, π‘’βˆ’ 1 bypass 𝑀 βˆ– {(𝑑+ 1, π‘’βˆ’ 1)},βˆ’) =

= 𝑄(𝐾) +𝑄(𝑑, 𝑒 bypass 𝑀 βˆ– {(𝑑, 𝑒)},+)+

+𝑄(𝑑, π‘’βˆ’ 1 bypass 𝑀 βˆ– {(𝑑, π‘’βˆ’ 1)}) +𝑄(𝑑+ 1, π‘’βˆ’ 1 bypass 𝑀 βˆ– {(𝑑+ 1, π‘’βˆ’ 1)},βˆ’) = 𝑄(𝑀 βˆͺ {(𝑑, π‘’βˆ’ 1)}).

Thus if the point (𝑑, π‘’βˆ’ 1) is added to the set 𝑀 then the probability 𝑄(𝑀) is not changed.This way we put new points onto the diagonal 𝑑+ 𝑒 = π‘šβˆ’ 1, therefore, the maximal diagonal is not changed.

Thus if we add a few points to the set 𝑀 , then the paths bypassing other points of the set 𝑀 bypass the points of𝑀 in the maximal diagonal as well. Therefore, if we remove all the points from the maximal diagonal, then 𝑄(𝑀)is not changed and no infinite path bypassing the points of the set 𝑀 appears. This way we change 𝑀 keeping𝑄(𝑀) fixed but decreasing the number of a maximal diagonal. By the inductive hypothesis the new 𝑄(𝑀) equals1, hence the old one also equals 1.

6. Answer. For each π‘₯ we have π‘Ž1(π‘₯, 𝑦) = π‘Ž1(βˆ’π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) + π‘Ž1(π‘₯, 𝑦) = π‘Ž2(2βˆ’ π‘₯, 𝑦) + π‘Ž1(2βˆ’ π‘₯, 𝑦).Solution. For each path 𝑠 denote by 𝑓(𝑠) the reflection of 𝑠 with respect to the 𝑦 axis. Then if 𝑠 is a path to

(π‘₯, 𝑦), then 𝑓(𝑠) is a path to (βˆ’π‘₯, 𝑦).For each path 𝑠 denote by 𝑔(𝑠) the path consisting of the same moves as 𝑠, but in the opposite order. Notice

that reordering of moves does not affect the endpoint.Now consider a path 𝑠 to (π‘₯, 𝑦) such that 𝑑(𝑠) is odd. Then the last move in 𝑠 is upwards-left. Therefore,

the last move in 𝑓(𝑠) is upwards-right, hence the first move in 𝑔(𝑓(𝑠)) is upwards-right. Thus 𝑓 ∘ 𝑔 is a bijectionbetween paths to (π‘₯, 𝑦) with odd number of tunings beginning with an upwards-right move and paths to (βˆ’π‘₯, 𝑦)with odd number of tunings beginning with an upwards-right move.

To prove the second equation use the result of Problem 4:

π‘Ž1(π‘₯, 𝑦) + π‘Ž2(π‘₯, 𝑦) =√2 Β· π‘Ž1(π‘₯βˆ’ 1, 𝑦 + 1) =

√2 Β· π‘Ž1(1βˆ’ π‘₯, 𝑦 + 1) = π‘Ž1(2βˆ’ π‘₯, 𝑦) + π‘Ž2(2βˆ’ π‘₯, 𝑦).

14

Remark. These identities can also be proved simultaneously by induction over 𝑦 using Problem 4. AlsoProblem 9 implies another identity (𝑦 βˆ’ π‘₯) π‘Ž2(π‘₯, 𝑦) = (𝑦 + π‘₯βˆ’ 2) π‘Ž2(2βˆ’ π‘₯, 𝑦).

7. Answer : for each 0 < 𝑦′ < 𝑦 we have

π‘Ž1(π‘₯, 𝑦) =βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž1(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1) + π‘Ž1(π‘₯β€², 𝑦′)π‘Ž2(π‘₯

β€² βˆ’ π‘₯+ 1, 𝑦 βˆ’ 𝑦′ + 1)],

π‘Ž2(π‘₯, 𝑦) =βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž2(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1)βˆ’ π‘Ž1(π‘₯β€², 𝑦′)π‘Ž1(π‘₯

β€² βˆ’ π‘₯+ 1, 𝑦 βˆ’ 𝑦′ + 1)].

The required expression for οΏ½οΏ½(π‘₯, 199) is obtained by taking 𝑦 = 199 and 𝑦′ = 100.Solution. Fix any positive integer 𝑦′ < 𝑦. Consider a path 𝑠 from (0, 0) to (π‘₯, 𝑦). Denote by (π‘₯β€², 𝑦′) the square

at which 𝑠 intersects the line 𝑦 = 𝑦′. Denote by 𝑠1 the part of 𝑠 that joins (0, 0) with (π‘₯β€², 𝑦′) and by 𝑠2 the partstarting at the intersection square of 𝑠 with the line 𝑦 = 𝑦′ βˆ’ 1 and ending at (π‘₯, 𝑦). Translate the path 𝑠2 so thatit starts at (0, 0).

Denote οΏ½οΏ½(𝑠) = (π‘Ž1(𝑠), π‘Ž2(𝑠)). Then

π‘Ž1(𝑠) =

{π‘Ž1(𝑠1)π‘Ž2(𝑠2) if the move to (π‘₯β€², 𝑦′) is upwards-left,

π‘Ž2(𝑠1)π‘Ž1(𝑠2) if the move to (π‘₯β€², 𝑦′) is upwards-right,

π‘Ž2(𝑠) =

{βˆ’π‘Ž1(𝑠1)π‘Ž1(𝑠2), if the move to (π‘₯β€², 𝑦′) is upwards-left,

π‘Ž2(𝑠1)π‘Ž2(𝑠2), if the move to (π‘₯β€², 𝑦′) is upwards-right.

Therefore,

π‘Ž1(π‘₯, 𝑦) =βˆ‘π‘ 

π‘Ž1(𝑠) =βˆ‘π‘₯β€²

βˆ‘π‘  through (π‘₯β€²,𝑦′)

π‘Ž1(𝑠) =βˆ‘π‘₯β€²

βŽ‘βŽ’βŽ’βŽ’βŽ’βŽ’βŽ£βˆ‘

𝑠 through (π‘₯β€²,𝑦′),move to (π‘₯β€²,𝑦′)is upwards-right

π‘Ž2(𝑠1)π‘Ž1(𝑠2) +βˆ‘

𝑠 through(π‘₯β€²,𝑦′),move to (π‘₯β€²,𝑦′)is upwards-left

π‘Ž1(𝑠1)π‘Ž2(𝑠2)

⎀βŽ₯βŽ₯βŽ₯βŽ₯βŽ₯⎦ =

=βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž1(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1) + π‘Ž1(π‘₯β€², 𝑦′)π‘Ž2(π‘₯

β€² βˆ’ π‘₯+ 1, 𝑦 βˆ’ 𝑦′ + 1)].

The formula for π‘Ž2(π‘₯, 𝑦) is proven analogously.Remark. Formulae illustrating the Huygens principle have the form of a convolution. In combinatorics the

convolution of two given sequences {π‘Žπ‘˜} and {𝑏𝑙} (π‘˜, 𝑙 β‰₯ 0) is the sequence {𝑐𝑛} (𝑛 β‰₯ 0), defined by the equality𝑐𝑛 =

βˆ‘π‘˜+𝑙=𝑛 π‘Žπ‘˜π‘π‘™. A convolution of two sequences corresponds to the product of their generating functions: if

𝐴(𝑑) =

βˆžβˆ‘π‘˜=0

π‘Žπ‘˜π‘‘π‘˜, 𝐡(𝑑) =

βˆžβˆ‘π‘™=0

𝑏𝑙𝑑𝑙, 𝐢(𝑑) =

βˆžβˆ‘π‘›=0

𝑐𝑛𝑑𝑛,

then 𝐢(𝑑) = 𝐴(𝑑)𝐡(𝑑). Therefore, choosing the generating functions in a right way, we can write down the Huygensprinciple in a more compact form.

Define two series of Laurent polynomials (i.e., polynomials in variables 𝑑 and π‘‘βˆ’1):

𝑃𝑛 = 𝑃𝑛(𝑑) =

π‘›βˆ‘π‘š=βˆ’π‘›+2

π‘Ž1(π‘š,𝑛)π‘‘π‘š, 𝑄𝑛 = 𝑄𝑛(𝑑) =

π‘›βˆ‘π‘š=βˆ’π‘›+2

π‘Ž2(π‘š,𝑛)π‘‘π‘š.

15

These polynomials can be considered as generating fuctions of the sequences π‘Ž1,2(π‘š,𝑛) with fixed 𝑛. From theanswer to the Problems 6 and 7 we have

π‘Ž2(π‘₯, 𝑦) =βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž2(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1)βˆ’ π‘Ž1(π‘₯β€², 𝑦′)π‘Ž1(π‘₯

β€² βˆ’ π‘₯+ 1, 𝑦 βˆ’ 𝑦′ + 1)]=

=βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž2(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1)βˆ’ π‘Ž1(π‘₯β€², 𝑦′)π‘Ž1(π‘₯βˆ’ π‘₯β€² βˆ’ 1, 𝑦 βˆ’ 𝑦′ + 1)

].

This formula is equivalent to the following addition theorem for 𝑄–polynomials:

𝑄𝑛+π‘š =1

π‘‘π‘„π‘›π‘„π‘š+1 βˆ’ π‘‘π‘ƒπ‘›π‘ƒπ‘š+1.

Also from the Problems 6 and 7 we have

π‘Ž1(π‘₯, 𝑦) =βˆ‘π‘₯β€²

[π‘Ž2(π‘₯

β€², 𝑦′)π‘Ž1(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1) + π‘Ž1(π‘₯β€², 𝑦′)π‘Ž2(π‘₯

β€² βˆ’ π‘₯+ 1, 𝑦 βˆ’ 𝑦′ + 1)]=

=βˆ‘π‘₯β€²

[π‘Ž2(π‘₯β€², 𝑦′)π‘Ž1(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1)+

+ π‘Ž1(π‘₯β€², 𝑦′)(π‘Ž2(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1) + π‘Ž1(π‘₯βˆ’ π‘₯β€² + 1, 𝑦 βˆ’ 𝑦′ + 1)βˆ’ π‘Ž1(π‘₯βˆ’ π‘₯β€² βˆ’ 1, 𝑦 βˆ’ 𝑦′ + 1))].

This formula respectively is equivalent to the following addition theorem for 𝑃–polynomials:

𝑃𝑛+π‘š =1

𝑑(π‘„π‘›π‘ƒπ‘š+1 + π‘ƒπ‘›π‘„π‘š+1) +

(1

π‘‘βˆ’ 𝑑

)π‘ƒπ‘›π‘ƒπ‘š+1.

In particular substitution of π‘š = 1 gives Dirac’s equation written in terms of 𝑃– and 𝑄–polynomials:{𝑃𝑛+1 = 1

𝑑𝑄𝑛𝑃2 + 𝑃𝑛 Β· (1𝑑𝑄2 +1𝑑𝑃2 βˆ’ 𝑑𝑃2) =

1π‘‘βˆš2(𝑄𝑛 + 𝑃𝑛),

𝑄𝑛+1 = 1𝑑𝑄𝑛𝑄2 βˆ’ 𝑑𝑃𝑛𝑃2 =

π‘‘βˆš2(𝑄𝑛 βˆ’ 𝑃𝑛).

It follows immediately that{π‘‘βˆš2𝑃𝑛+1 βˆ’π‘„π‘› = 𝑃𝑛 = 𝑄𝑛 βˆ’

√2𝑑 𝑄𝑛+1

π‘‘βˆš2𝑃𝑛+1 βˆ’ 𝑃𝑛 = 𝑄𝑛 = 𝑃𝑛 +

√2𝑑 𝑄𝑛+1

β‡’

{𝑃𝑛+1 =

√2𝑑 𝑄𝑛 βˆ’ 𝑄𝑛+1

𝑑2

𝑄𝑛+1 = 𝑑2𝑃𝑛+1 βˆ’ π‘‘βˆš2𝑃𝑛

β‡’

{𝑃𝑛+1 = 1√

2(𝑑+ 1

𝑑 )𝑃𝑛 βˆ’ π‘ƒπ‘›βˆ’1

𝑄𝑛+1 = 1√2(𝑑+ 1

𝑑 )𝑄𝑛 βˆ’π‘„π‘›βˆ’1.

The latter formulae give us yet another recursive definitions for 𝑃𝑛 and 𝑄𝑛.

8. Answer : the graphs of 𝑃 (π‘₯, 1000), π‘Ž1(π‘₯, 1000), and π‘Ž2(π‘₯, 1000) respectively as functions in π‘₯:

-1000 -500 500 1000

0.0005

0.0010

0.0015

0.0020

0.0025

0.0030

-1000 -500 500 1000

-0.06

-0.04

-0.02

0.02

0.04

0.06

-1000 -500 500 1000

-0.05

0.05

In the above graph of 𝑃 (π‘₯, 1000), large values are cut out, to make the oscillations more visible. Compare withthe following graphs of 𝑃 (π‘₯, 1000) and 𝑃 (π‘₯, 𝑦) without cutting (kindly provided by A. Daniyarkhodzhaev andF. Kuyanov, participants of the Summer School in Contemporary Mathematics in Dubna, 2019):

16

9. Answer 1: for each 𝑦 > |π‘₯| we have

π‘Ž1(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2βˆ‘

𝑀≑2π‘¦βˆ’π‘₯βˆ’2

2

(βˆ’1)π‘¦βˆ’π‘₯βˆ’2+2𝑀·sign(π‘₯)

4

(|π‘₯|𝑀

)( π‘¦βˆ’|π‘₯|βˆ’22

π‘¦βˆ’π‘₯βˆ’2βˆ’2𝑀4

),

π‘Ž2(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2βˆ‘

𝑀≑2π‘¦βˆ’π‘₯2

(βˆ’1)π‘¦βˆ’π‘₯+2𝑀·sign(π‘₯)

4

(|π‘₯|𝑀

)( π‘¦βˆ’|π‘₯|βˆ’22

π‘¦βˆ’π‘₯βˆ’2𝑀4

);

for each 𝑦 = π‘₯ > 0 we have

π‘Ž1(𝑦, 𝑦) = 0,

π‘Ž2(𝑦, 𝑦) = 2(1βˆ’π‘¦)/2;

for each 0 < 𝑦 < |π‘₯| or 𝑦 = βˆ’π‘₯ > 0 we have

π‘Ž1(π‘₯, 𝑦) = 0,

π‘Ž2(π‘₯, 𝑦) = 0.

Answer 2: for each 𝑦 > |π‘₯| we have

π‘Ž1(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ((π‘₯+ 𝑦 βˆ’ 2)/2

π‘Ÿ

)((𝑦 βˆ’ π‘₯βˆ’ 2)/2

π‘Ÿ

),

π‘Ž2(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=1

(βˆ’1)π‘Ÿ((π‘₯+ 𝑦 βˆ’ 2)/2

π‘Ÿ

)((𝑦 βˆ’ π‘₯βˆ’ 2)/2

π‘Ÿ βˆ’ 1

);

For other 𝑦 the answer is the same as Answer 1.Answer 3. For even 𝑝+ π‘ž and 2βˆ’ π‘ž ≀ 𝑝 ≀ π‘ž we have25

π‘Ž1(𝑝, π‘ž) = (βˆ’1)π‘ž2(π‘žβˆ’3)/2 Β·π‘žβˆ’2βˆ‘

π‘˜=π‘ž+|𝑝|βˆ’2

2

(βˆ’2)βˆ’π‘˜ Β· π‘˜!

(π‘ž βˆ’ 2βˆ’ π‘˜)!(π‘βˆ’π‘ž2 + 1 + π‘˜)!(π‘˜ βˆ’ 𝑝+π‘ž

2 + 1)!

π‘Ž2(𝑝, π‘ž) = 2(1βˆ’π‘ž)/2 Β·(π‘ž βˆ’ 1π‘žβˆ’π‘2

)βˆ’ (βˆ’1)π‘ž2(π‘žβˆ’3)/2

π‘žβˆ’2βˆ‘π‘˜=

π‘ž+|𝑝|βˆ’22

(βˆ’2)βˆ’π‘˜ Β· π‘˜!(π‘˜ + 𝑝+ 1)

(π‘ž βˆ’ 1βˆ’ π‘˜)!(π‘βˆ’π‘ž2 + 1 + π‘˜)!(π‘˜ βˆ’ 𝑝+π‘ž

2 + 1)!

Remark. Answer 2 shows that π‘Ž1(βˆ’π‘¦+2π‘˜, 𝑦) and π‘Ž2(βˆ’π‘¦+2π‘˜, 𝑦) are the coefficients before π‘‘π‘¦βˆ’π‘˜βˆ’1 and π‘‘π‘¦βˆ’π‘˜ inthe expansion of 2(1βˆ’π‘¦)/2(1 + 𝑑)π‘¦βˆ’π‘˜βˆ’1(1βˆ’ 𝑑)π‘˜βˆ’1 for 1 ≀ π‘˜ ≀ 𝑦 βˆ’ 1. In particular,

π‘Ž1(0, 4𝑛+ 2) =(βˆ’1)𝑛

2(4𝑛+1)/2

(2𝑛

𝑛

),

π‘Ž2(0, 4𝑛+ 2) = 0,

and

π‘Ž1(0, 4𝑛) = 0,

π‘Ž2(0, 4𝑛) =(βˆ’1)𝑛

2(4π‘›βˆ’1)/2

(2π‘›βˆ’ 1

𝑛

).

The sums in the expressions for π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) for 𝑦 > |π‘₯| are particular cases of well-known Gaussianhypergeometric functions (but familiarity with them is not required for what follows):

π‘Ž1(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/22𝐹1

(1βˆ’ π‘₯+ 𝑦

2, 1 +

π‘₯βˆ’ 𝑦

2, 1;βˆ’1

),

π‘Ž2(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2

(1βˆ’ π‘₯+ 𝑦

2

)2𝐹1

(2βˆ’ π‘₯+ 𝑦

2, 1 +

π‘₯βˆ’ 𝑦

2, 2;βˆ’1

).

Hint. Let us find π‘Ž1(π‘₯, 𝑦) for 𝑦 > |π‘₯|. Consider a path from the square (0, 0) to the square (π‘₯, 𝑦) with anodd number of turns (the others do not contribute to π‘Ž1(π‘₯, 𝑦)). Denote by 𝑅 and 𝐿 the number of upwards-right

25Fractions in those equalities are respectively(π‘˜

π‘ž βˆ’ 2βˆ’ π‘˜, π‘βˆ’π‘ž2 + 1 + π‘˜, π‘˜ βˆ’ 𝑝+π‘ž

2 + 1

)and

(π‘˜

π‘ž βˆ’ 2βˆ’ π‘˜, π‘βˆ’π‘ž2 + 1 + π‘˜, π‘˜ βˆ’ 𝑝+π‘ž

2 + 1

)Β· π‘˜ + 𝑝+ 1

π‘ž βˆ’ 1βˆ’ π‘˜.

17

and upwards-left moves respectively. These numbers are uniquely determined by the conditions 𝐿 + 𝑅 = 𝑦 andπ‘…βˆ’ 𝐿 = π‘₯. Denote by 2π‘Ÿ + 1 the number of turns in the path. Let π‘₯1, π‘₯2, . . . , π‘₯π‘Ÿ+1 be the number of consecutiveupwards-right moves before the first, the third, . . . , the last turn respectively. Let 𝑦1, 𝑦2, . . . , π‘¦π‘Ÿ+1 be the numberof consecutive upwards-left moves after the first, the third, . . . , the last turn respectively. Then π‘₯π‘˜, π‘¦π‘˜ β‰₯ 1 and

𝑅 = π‘₯1 + Β· Β· Β·+ π‘₯π‘Ÿ+1;

𝐿 = 𝑦1 + Β· Β· Β·+ π‘¦π‘Ÿ+1.

Conversely, each collection (π‘₯1, π‘₯2, . . . , π‘₯π‘Ÿ+1, 𝑦1, 𝑦2, . . . , π‘¦π‘Ÿ+1) satisfying the resulting equations determines a pathfrom (0, 0) to (π‘₯, 𝑦) with an odd number of turns.

The problem now reduces to a combinatorial one: find the number of positive integer solutions of the resultingequations. For the first equation, this number equals to the number of ways to put π‘Ÿ sticks between 𝑅 coins in arow, that is,

(π‘…βˆ’1π‘Ÿ

). Thus

π‘Ž1(π‘₯, 𝑦) = 2(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ(π‘…βˆ’ 1

π‘Ÿ

)(πΏβˆ’ 1

π‘Ÿ

).

Since 𝐿+𝑅 = 𝑦 and π‘…βˆ’ 𝐿 = π‘₯, the required formula from Answer 2 follows.The formula for π‘Ž1(π‘₯, 𝑦) from Answer 1 can be derived using the polynomial from the remark after the answer.

Depending on the sign of π‘₯ we get

2(1βˆ’π‘¦)/2(1 + 𝑑)π‘¦βˆ’π‘˜βˆ’1(1βˆ’ 𝑑)π‘˜βˆ’1 =

{2(1βˆ’π‘¦)/2(1βˆ’ 𝑑2)

π‘¦βˆ’π‘₯βˆ’22 (1βˆ’ 𝑑)π‘₯, for π‘₯ β‰₯ 0;

2(1βˆ’π‘¦)/2(1βˆ’ 𝑑2)𝑦+π‘₯βˆ’2

2 (1 + 𝑑)βˆ’π‘₯, for π‘₯ < 0.

For each |π‘₯| < 𝑦 the numbers π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) are equal to the coefficients before 𝑑π‘₯βˆ’π‘¦βˆ’2

2 and 𝑑π‘₯βˆ’π‘¦2 in the

expansion of the polynomials. Considering the cases π‘₯ β‰₯ 0 and π‘₯ < 0 separately we get the required formula.The proof of the formulae for π‘Ž2(π‘₯, 𝑦) for 𝑦 > |π‘₯| is analogous. The case 𝑦 ≀ |π‘₯| is obvious.

Solution by Alexander Kudryavtsev (participant of Summer conference of Tournament of Towns). Denote𝑐𝑝,π‘ž = π‘Ž1(2𝑝 βˆ’ π‘ž + 1, π‘ž + 1) Β· 2π‘ž/2, 𝑑𝑝,π‘ž = π‘Ž2(2𝑝 βˆ’ π‘ž + 1, π‘ž + 1) Β· 2π‘ž/2 (they are well-defined for 0 ≀ 𝑝 ≀ π‘ž βˆ’ 1). Let𝐢(π‘₯, 𝑦), 𝐷(π‘₯, 𝑦) be generating functions of 𝑐𝑝,π‘ž and 𝑑𝑝,π‘ž respectively. From the answer to Problem 4 we obtain thefollowing equations: ⎧βŽͺ⎨βŽͺ⎩

𝐢(π‘₯, 𝑦)βˆ’ 𝐢(π‘₯, 0)

𝑦= 𝐷(π‘₯, 𝑦) + 𝐢(π‘₯, 𝑦)

𝐷(π‘₯, 𝑦)βˆ’π·(π‘₯, 0)

𝑦= π‘₯ Β· (𝐷(π‘₯, 𝑦)βˆ’ 𝐢(π‘₯, 𝑦))

Since 𝐢(π‘₯, 0) = 0 and 𝐷(π‘₯, 0) = 1, we get{𝐢(π‘₯, 𝑦) = 𝑦 Β· (𝐷(π‘₯, 𝑦) + 𝐢(π‘₯, 𝑦))

𝐷(π‘₯, 𝑦)βˆ’ 1 = π‘₯𝑦 Β· (𝐷(π‘₯, 𝑦)βˆ’ 𝐢(π‘₯, 𝑦))

The solution of this system is 𝐢(π‘₯, 𝑦) =𝑦

1βˆ’ 𝑦 βˆ’ π‘₯𝑦 + 2π‘₯𝑦2and 𝐷(π‘₯, 𝑦) =

1βˆ’ 𝑦

1βˆ’ 𝑦 βˆ’ π‘₯𝑦 + 2π‘₯𝑦2. Denote

𝐸(π‘₯, 𝑦) :=𝐢(π‘₯, 𝑦)

𝑦=

𝐷(π‘₯, 𝑦)

1βˆ’ 𝑦= 1 + 𝑦(1 + π‘₯βˆ’ 2π‘₯𝑦) + 𝑦2(1 + π‘₯βˆ’ 2π‘₯𝑦)2 + . . .

The coefficient of π‘₯𝑛 Β· π‘¦π‘š in the 𝐸(π‘₯, 𝑦) is equal to

π‘šβˆ‘π‘˜=max(𝑛,π‘šβˆ’π‘›)

(βˆ’2)π‘šβˆ’π‘˜ Β· π‘˜!

(π‘šβˆ’ π‘˜)!(π‘›βˆ’π‘š+ π‘˜)!(π‘˜ βˆ’ 𝑛)!,

since from every summand of the form π‘¦π‘˜(1 + π‘₯βˆ’ 2π‘₯𝑦)π‘˜ we must take exactly one combination of factors:

βˆ™ for the power of 𝑦 to be equal to π‘š, the number of factors βˆ’2π‘₯𝑦 must be exactly π‘šβˆ’ π‘˜;

βˆ™ for the power of π‘₯ to be equal to 𝑛, the number of factors π‘₯ must be exactly π‘›βˆ’ (π‘šβˆ’ π‘˜);

βˆ™ for the total number of factors to be π‘˜, the number of factors 1 must be π‘˜βˆ’ (π‘›βˆ’ (π‘šβˆ’π‘˜))βˆ’ (π‘šβˆ’π‘˜) = π‘˜βˆ’π‘›.

18

And the number of ways to choose π‘š βˆ’ π‘˜, π‘˜ βˆ’ 𝑛 and 𝑛 βˆ’ (π‘š βˆ’ π‘˜) factors (1 + π‘₯ βˆ’ 2π‘₯𝑦) equals to the fraction inthe formula for 𝐸(π‘₯, 𝑦).

Now Answer 3 can be easily obtained from the equations 𝐢(π‘₯, 𝑦) = 𝑦 ·𝐸(π‘₯, 𝑦) and 𝐷(π‘₯, 𝑦) = 𝐸(π‘₯, 𝑦)βˆ’πΆ(π‘₯, 𝑦).

10. Answer:

οΏ½οΏ½(π‘₯, 𝑦) =

√2(sin

(πœ‹π‘¦4 βˆ’ π‘₯2

2𝑦

), 𝑒π‘₯/𝑦 cos

(πœ‹π‘¦4 + π‘₯(2βˆ’π‘₯)

2𝑦

))βˆšπœ‹π‘¦

+O

(log2 𝑦

𝑦

)for |π‘₯| = O(

βˆšπ‘¦) .

Remark. Compare the graphs of 𝑃 (π‘₯, 400), π‘Ž1(π‘₯, 400), and π‘Ž2(π‘₯, 400) (blue) with the graphs of their approxi-mations in the right-hand side for 𝑦 = 400 (orange):

-400 -200 200 400

0.001

0.002

0.003

0.004

0.005

0.006

0.007

-400 -200 200 400

-0.05

0.05

0.10

-400 -200 200 400

-0.10

-0.05

0.05

0.10

Hint. Note that by Problem 9

π‘Ž2(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) = 2βˆ’π‘›/2[π‘₯π‘˜βˆ’1](βˆ’1)π‘˜(1 + π‘₯)π‘›βˆ’π‘˜βˆ’1(1βˆ’ π‘₯)π‘˜;

π‘Ž1(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) = 2βˆ’π‘›/2[π‘₯π‘˜](βˆ’1)π‘˜(1 + π‘₯)π‘›βˆ’π‘˜βˆ’1(1βˆ’ π‘₯)π‘˜.

Thus

π‘Ž2(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) = 2βˆ’π‘›/2(βˆ’1)π‘˜βˆ« 1

0(1 + 𝑒2πœ‹π‘–π‘‘)π‘›βˆ’π‘˜βˆ’1(1βˆ’ 𝑒2πœ‹π‘–π‘‘)π‘˜π‘’βˆ’2πœ‹π‘–(π‘˜βˆ’1)𝑑𝑑𝑑.

For an odd 𝑛 denote 𝑙 = π‘˜ βˆ’ π‘›βˆ’12 . We have

π‘Ž2(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) =π‘–π‘˜βˆš2

∫ 1

0(sin(2πœ‹π‘‘))

π‘›βˆ’12 (tan(πœ‹π‘‘))π‘™π‘’βˆ’2πœ‹π‘–(π‘™βˆ’1)𝑑𝑑𝑑.

Since 𝑙 << 𝑛, we can calculate the integral in the areas (1/4βˆ’ 𝛿, 1/4 + 𝛿), (3/4βˆ’ 𝛿, 3/4 + 𝛿) with good accuracy.Denote π‘ˆ β‰ͺ

βˆšπ‘› log 𝑛, |𝑙| ≀ π‘ˆ.

In (1/4βˆ’ 𝛿, 1/4 + 𝛿) we have 𝑑 = 1/4 + 𝜏, |𝜏 | ≀ 𝛿,

sin(2πœ‹π‘‘) = cos(2πœ‹πœ) = 1βˆ’ 2πœ‹2𝜏2 +𝑂(𝜏4) = π‘’βˆ’2πœ‹2𝜏2 +𝑂(𝜏4);

tan(πœ‹π‘‘) = 1 + 2πœ‹πœ +𝑂(𝜏2) = 𝑒2πœ‹πœ +𝑂(𝜏2);

π‘’βˆ’2πœ‹π‘–(π‘™βˆ’1)𝑑 = 𝑖1βˆ’π‘™π‘’βˆ’2πœ‹π‘–(π‘™βˆ’1)𝜏

and we getπ‘–π‘˜+1βˆ’π‘™

√2

∫ 𝛿

βˆ’π›Ώ(π‘’βˆ’2πœ‹2𝜏2 +𝑂(𝜏4))

π‘›βˆ’12 (𝑒2πœ‹πœ +𝑂(𝜏2))π‘™π‘’βˆ’2πœ‹π‘–(π‘™βˆ’1)πœπ‘‘πœ. (*)

As π‘›πœ4 β‰ͺ 1 and π‘™πœ2 β‰ͺ 1, i.e. 𝑛𝛿4 β‰ͺ 1 and π‘ˆπ›Ώ2 β‰ͺ 1,

(*) = 𝑖𝑛+12

√2

∫ 𝛿

βˆ’π›Ώ(π‘’βˆ’(π‘›βˆ’1)πœ‹2𝜏2 +𝑂(𝜏4))(𝑒2πœ‹πœπ‘™ +𝑂(𝜏2))π‘’βˆ’2πœ‹π‘–(π‘™βˆ’1)πœπ‘‘πœ.

If we open the brackets we get the summand 𝑂(π‘›πœ4𝑒2πœ‹π‘™πœ ) = 𝑂(𝑛𝛿4𝑒2πœ‹π‘ˆπ›Ώ) = 𝑂(π‘›βˆ’Ξ”), since we ask π‘ˆπ›Ώ β‰ͺ log 𝑛.Then

(*) = 𝑖𝑛+12

√2

∫ 𝛿

βˆ’π›Ώ(π‘’βˆ’(π‘›βˆ’1)πœ‹2𝜏2+2πœ‹πœπ‘™βˆ’2πœ‹π‘–(π‘™βˆ’1)𝜏 +𝑂(π‘›βˆ’Ξ”) +𝑂(π‘ˆπ›Ώ2))π‘‘πœ.

Now estimate the error∫ ∞

π›Ώπ‘’βˆ’(π‘›βˆ’1)πœ‹2𝜏2+2πœ‹πœπ‘™π‘‘πœ =

∫ ∞

π›Ώπ‘’βˆ’(π‘›βˆ’1)πœ‹2(πœβˆ’ 𝑙

πœ‹(π‘›βˆ’1))2+ 𝑙2

π‘›βˆ’1π‘‘πœ β‰ͺ∫ ∞

π›Ώπ‘’βˆ’(π‘›βˆ’1)πœ‹2 𝜏2

4+ 𝑙2

π‘›βˆ’1π‘‘πœ ≀ 1βˆšπ‘›βˆ’ 1

π‘’βˆ’(π‘›βˆ’1)𝛿2 ,

since 𝜏2 β‰₯ 𝑙

πœ‹(π‘›βˆ’1) , i.e. 𝛿 β‰₯ π‘ˆ/(π‘›βˆ’ 1).

19

So the necessary condition is 𝑛𝛿2 ≫ log 𝑛. Then we can calculate

(*) = 𝑖𝑛+12

√2

∫ ∞

βˆ’βˆžπ‘’βˆ’(π‘›βˆ’1)πœ‹2𝜏2+2πœ‹πœπ‘™βˆ’2πœ‹π‘–(π‘™βˆ’1)πœπ‘‘πœ+𝑂(π‘›βˆ’Ξ”)+𝑂(π‘ˆπ›Ώ3) =

𝑖𝑛+12

√2

∫ ∞

βˆ’βˆžπ‘’βˆ’(π‘›βˆ’1)πœ‹2𝜏2+2πœ‹πœπ‘™βˆ’2πœ‹π‘–(π‘™βˆ’1)πœπ‘‘πœ+𝑂(log2 𝑛/𝑛),

since π‘ˆπ›Ώ3 β‰ͺ log 𝑛 Β· 𝛿2 β‰ͺ log2 𝑛/𝑛.Finally, from the formula

βˆ«βˆžβˆ’βˆž π‘’βˆ’π‘Žπ‘‘2+𝑏𝑑𝑑𝑑 =

βˆšπœ‹/π‘Žπ‘’π‘

2/4π‘Ž we get

(*) = 𝑖𝑛+12√

2πœ‹(π‘›βˆ’ 1)𝑒

(2π‘™βˆ’1)βˆ’2𝑖(π‘™βˆ’1)π‘›βˆ’1 +𝑂(log2 𝑛/𝑛).

In the area of 3/4 the integral approximately equals 𝑖3𝑛+12√2

𝑒(2π‘™βˆ’1)+2𝑖(π‘™βˆ’1)

π‘›βˆ’1 and, since 3𝑛+12 ≑ βˆ’π‘›+1

2 mod 4,

π‘Ž2(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) = 𝑒2π‘™βˆ’1

√2

πœ‹(π‘›βˆ’ 1)cos

(πœ‹(𝑛+ 1)

4βˆ’ 2𝑙2

π‘›βˆ’ 1+βˆ’ 2𝑙

π‘›βˆ’ 1

)+𝑂(log2 𝑛/𝑛) =

= βˆ’π‘’2π‘™βˆ’1

√2

πœ‹(π‘›βˆ’ 1)sin

(πœ‹(π‘›βˆ’ 1)

4βˆ’ 2𝑙2

π‘›βˆ’ 1+

2𝑙

π‘›βˆ’ 1

)+𝑂(log2 𝑛/𝑛).

Analogously,

π‘Ž1(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) =

√2

πœ‹(π‘›βˆ’ 1)cos

(πœ‹(π‘›βˆ’ 1)

4βˆ’ 2𝑙2

π‘›βˆ’ 1

)+𝑂(log2 𝑛/𝑛).

For an even 𝑛 the answer is

π‘Ž2(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) = 𝑒2𝑙

√2

πœ‹(π‘›βˆ’ 2)cos

(πœ‹(𝑛+ 1)

4βˆ’ 2(𝑙2 βˆ’ 1/4)

π‘›βˆ’ 2

)+𝑂(log2 𝑛/𝑛);

π‘Ž1(βˆ’π‘›+ 2π‘˜ + 1, 𝑛+ 1) =

√2

πœ‹(π‘›βˆ’ 2)sin

(πœ‹(𝑛+ 1)

4βˆ’ 2(𝑙 + 1/2)2

π‘›βˆ’ 2

)+𝑂(log2 𝑛/𝑛),

where 𝑙 = π‘˜ βˆ’ 𝑛/2.

11. Answer : 𝑃 (π‘₯, 𝑦) = 𝑃 (π‘₯, 𝑦,+) + 𝑃 (π‘₯, 𝑦,βˆ’); οΏ½οΏ½(π‘₯, 𝑦,+) = (0, π‘Ž2(π‘₯, 𝑦)); οΏ½οΏ½(π‘₯, 𝑦,βˆ’) = (π‘Ž1(π‘₯, 𝑦), 0).Solution. Note that if a path 𝑠 from (0, 0) to (π‘₯, 𝑦) starts and finishes with an upward-left move, then the checker

changes the direction an odd number of times. Therefore, οΏ½οΏ½(𝑠) = (Β±21βˆ’π‘¦2 , 0). Analogously, if a path 𝑠 finishes with

an upward-right move, then οΏ½οΏ½(𝑠) = (0,Β±21βˆ’π‘¦2 ). Hence, οΏ½οΏ½(π‘₯, 𝑦,+) = (0, π‘Ž2(π‘₯, 𝑦)) and οΏ½οΏ½(π‘₯, 𝑦,βˆ’) = (π‘Ž1(π‘₯, 𝑦), 0).

For the probability, the following formula comprehensively explains the answer:

𝑃 (π‘₯, 𝑦) = |π‘Ž(π‘₯, 𝑦,+) + οΏ½οΏ½(π‘₯, 𝑦,βˆ’)|2 = π‘Ž21(π‘₯, 𝑦) + π‘Ž22(π‘₯, 𝑦) = 𝑃 (π‘₯, 𝑦,+) + 𝑃 (π‘₯, 𝑦,βˆ’).

12. Answer: for each integer 𝑦 β‰₯ 2 we have 𝑃 (𝑦,βˆ’) = 12

βˆ‘βŒŠπ‘¦/2βŒ‹βˆ’1π‘˜=0

1(βˆ’4)π‘˜

(2π‘˜π‘˜

). We have also max𝑦 𝑃 (𝑦,βˆ’) =

𝑃 (2,βˆ’) = 𝑃 (3,βˆ’) = 1/2 and limπ‘¦β†’βˆž 𝑃 (𝑦,βˆ’) = 12√2.

Remark. The graph of the sequence kindly provided by Gleb Minaev and Ivan Russkikh, participants ofSummer Conference of Tournament of Towns:

Solution. Denote

𝑆1(𝑦) =βˆ‘π‘₯

π‘Ž21(π‘₯, 𝑦); 𝑆2(𝑦) =βˆ‘π‘₯

π‘Ž22(π‘₯, 𝑦); 𝑆12(𝑦) =βˆ‘π‘₯

π‘Ž1(π‘₯, 𝑦)π‘Ž2(π‘₯, 𝑦).

By Problem 5 for each 𝑦 β‰₯ 1 we have𝑆1(𝑦) + 𝑆2(𝑦) = 1.

20

By the answers to Problem 4, Problem 6 and Problem 7 we have

π‘Ž1(0, 2𝑦) =1√2

βˆ‘π‘₯

π‘Ž1(π‘₯, 𝑦)(π‘Ž2(π‘₯, 𝑦)βˆ’ π‘Ž1(π‘₯, 𝑦)) + π‘Ž2(π‘₯, 𝑦)(π‘Ž2(π‘₯, 𝑦) + π‘Ž1(π‘₯, 𝑦)) =1√2(𝑆2(𝑦) + 2𝑆12(𝑦)βˆ’ 𝑆1(𝑦)).

By definition and the answer to Problem 4 we have

𝑆1(𝑦 + 1)βˆ’ 𝑆2(𝑦 + 1) = 2𝑆12(𝑦).

Hence,𝑆1(𝑦 + 1)βˆ’ 𝑆2(𝑦 + 1) = 𝑆1(𝑦)βˆ’ 𝑆2(𝑦) + π‘Ž1(0, 2𝑦)

√2.

Since 𝑆1(𝑦) + 𝑆2(𝑦) = 1, we have the recurrence relation 𝑆1(𝑦 + 1) = 𝑆1(𝑦) +1√2π‘Ž1(0, 2𝑦).

The remark from the solution of Problem 9 implies by induction that

𝑆1(𝑦) =1

2

βŒŠπ‘¦/2βŒ‹βˆ’1βˆ‘π‘˜=0

1

(βˆ’4)π‘˜

(2π‘˜

π‘˜

).

It only remains to recall that 𝑃 (𝑦,βˆ’) = 𝑆1(𝑦) by Problem 11.Since the sequence

(2𝑗𝑗

)14𝑗

decreases, it follows that 𝑃 (𝑦,βˆ’) < 𝑃 (3,βˆ’) for each 𝑦 > 3, thus max𝑦 𝑃 (𝑦,βˆ’) =𝑃 (2,βˆ’) = 𝑃 (3,βˆ’) = 1/2.

By the Newton binomial theorem we getβˆ‘βˆž

π‘˜=0

(2π‘˜π‘˜

)π‘₯π‘˜ = 1√

1βˆ’4π‘₯for each π‘₯ ∈

[βˆ’1

4 ,14

). Setting π‘₯ = βˆ’1

4 we

obtain limπ‘¦β†’βˆž 𝑃 (𝑦,βˆ’) = limπ‘¦β†’βˆž12

βˆ‘βˆžπ‘˜=0

(2π‘˜π‘˜

) (βˆ’1

4

)π‘˜= 1

2√2.

Remark. Using the Stirling formula (see Β§2) one can estimate the convergence rate:

|𝑃 (π‘˜,βˆ’)βˆ’ limπ‘¦β†’βˆž

𝑃 (𝑦,βˆ’)| < 1

2 Β· 4βŒŠπ‘˜/2βŒ‹

(2βŒŠπ‘˜/2βŒ‹βŒŠπ‘˜/2βŒ‹

)<

𝑒

2πœ‹βˆš

2βŒŠπ‘˜/2βŒ‹<

1

2βˆšπ‘˜.

13. Answer :

𝑃 (π‘₯, 𝑦, 0,+) =

{1, for π‘₯ = 𝑦;

0, for π‘₯ = 𝑦.

𝑃 (π‘₯, 𝑦,∞,+) =

{1, for π‘₯ = 1, 𝑦 ≑ 1 mod 2;

0, otherwise.

Solution. By the definition, if the number of turns in a path 𝑠 is nonzero then οΏ½οΏ½(𝑠, 0) = 0. Therefore,οΏ½οΏ½(π‘₯, 𝑦, 0,+) = 0 if and only if π‘₯ = 𝑦. Hence, we get the answer for 𝑃 (π‘₯, 𝑦, 0,+). The case π‘₯ = 𝑦 is obvious.

One cannot prove the answer for 𝑃 (π‘₯, 𝑦,∞,+) because it is a definition; but let us provide a motivation forthe definition. If π‘šπœ€ is β€œvery large”, then οΏ½οΏ½(𝑠,π‘šπœ€) is β€œsmall” unless the checker turns at each move; this gives theformula for 𝑃 (π‘₯, 𝑦,∞,+).

Remark. 𝑃 (π‘₯, 𝑦,∞,+) = limπ‘šπœ€β†’βˆž 𝑃 (π‘₯, 𝑦,π‘šπœ€,+). Indeed,

limπ‘šπœ€β†’βˆž

|π‘Ž(𝑠,π‘šπœ€)| = limπ‘šπœ€β†’βˆž

(π‘šπœ€)𝑑(𝑠)

(√1 + (π‘šπœ€)2)π‘¦βˆ’1

= limπ‘šπœ€β†’βˆž

1

(√1 + (π‘šπœ€)2)π‘¦βˆ’1βˆ’π‘‘(𝑠)

=

{1, for 𝑑(𝑠) = 𝑦 βˆ’ 1;

0, otherwise;

where 𝑑(𝑠) is the number of turns in the path 𝑠. To compute 𝑃 (π‘₯, 𝑦,π‘šπœ€,+), we take only paths with 𝑑 ≑ 0 mod 2,hence we get the answer.

14. Answer. Recall the notation introduced at the beginning of section β€œHints. . . ”. We have

π‘Ž1(π‘₯, 𝑦,π‘šπœ€,βˆ’) =1√

1 +π‘š2πœ€2(π‘Ž1(π‘₯+ 1, 𝑦 βˆ’ 1,π‘šπœ€,βˆ’) +π‘šπœ€π‘Ž2(π‘₯+ 1, 𝑦 βˆ’ 1,π‘šπœ€,+)),

π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+) =1√

1 +π‘š2πœ€2(π‘Ž2(π‘₯βˆ’ 1, 𝑦 βˆ’ 1,π‘šπœ€,+)βˆ’π‘šπœ€π‘Ž1(π‘₯βˆ’ 1, 𝑦 βˆ’ 1,π‘šπœ€,βˆ’)),

π‘Ž2(π‘₯, 𝑦,π‘šπœ€,βˆ’) = π‘Ž1(π‘₯, 𝑦,π‘šπœ€,+) = 0.

For each 𝑦 > |π‘₯| we have

π‘Ž1(π‘₯, 𝑦,π‘šπœ€) = (1 +π‘š2πœ€2)(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ((π‘₯+ 𝑦 βˆ’ 2)/2

π‘Ÿ

)((𝑦 βˆ’ π‘₯βˆ’ 2)/2

π‘Ÿ

)(π‘šπœ€)2π‘Ÿ+1,

π‘Ž2(π‘₯, 𝑦,π‘šπœ€) = (1 +π‘š2πœ€2)(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=1

(βˆ’1)π‘Ÿ((π‘₯+ 𝑦 βˆ’ 2)/2

π‘Ÿ

)((𝑦 βˆ’ π‘₯βˆ’ 2)/2

π‘Ÿ βˆ’ 1

)(π‘šπœ€)2π‘Ÿ;

21

For each 𝑦 = π‘₯ > 0 we have

π‘Ž1(𝑦, 𝑦,π‘šπœ€) = 0,

π‘Ž2(𝑦, 𝑦,π‘šπœ€) = (1 +π‘š2πœ€2)(1βˆ’π‘¦)/2;

For each 0 < 𝑦 < |π‘₯| or 𝑦 = βˆ’π‘₯ > 0 we have

π‘Ž1(π‘₯, 𝑦,π‘šπœ€) = 0,

π‘Ž2(π‘₯, 𝑦,π‘šπœ€) = 0.

Hint. First let us solve the analogue of Problem 4 for π‘šπœ€ = 1. Let us derive the formula for π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+).Consider any path 𝑠 from (0, 0) to (π‘₯, 𝑦).

The last move in the path 𝑠 is made either from (π‘₯βˆ’ 1, 𝑦) or from (π‘₯+ 1, 𝑦). Denote by 𝑠′ the path 𝑠 withoutthe last move. If the directions of the last moves in 𝑠 and 𝑠′ coincide, then οΏ½οΏ½(𝑠,π‘šπœ€) = 1√

1+π‘š2πœ€2οΏ½οΏ½(𝑠′,π‘šπœ€), otherwise

οΏ½οΏ½(𝑠,π‘šπœ€) = π‘šπœ€βˆš1+π‘š2πœ€2

(π‘Ž2(𝑠′,π‘šπœ€),βˆ’π‘Ž1(𝑠

β€²,π‘šπœ€)).

Therefore, π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+) = 1√1+π‘š2πœ€2

(π‘Ž2(π‘₯βˆ’ 1, 𝑦 βˆ’ 1,π‘šπœ€,+)βˆ’π‘šπœ€π‘Ž1(π‘₯βˆ’ 1, 𝑦 βˆ’ 1,π‘šπœ€,βˆ’)). The formula for

π‘Ž1(π‘₯, 𝑦,π‘šπœ€,βˆ’) is proved analogously.Now let us solve the analogue of Problem 5: we prove that

βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦,π‘šπœ€) = 1 for all 𝑦 β‰₯ 1 by induction over

𝑦 using these results. Obviously,βˆ‘π‘₯∈Z

𝑃 (π‘₯, 1,π‘šπœ€) = 1. The step of induction follows immediately from the following

computation:βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦+1,π‘šπœ€) =βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦 + 1,π‘šπœ€,βˆ’)2 + π‘Ž2(π‘₯, 𝑦 + 1,π‘šπœ€,+)2

]=

βˆ‘π‘₯∈Z

π‘Ž1(π‘₯, 𝑦+1,π‘šπœ€,βˆ’)2+βˆ‘π‘₯∈Z

π‘Ž2(π‘₯, 𝑦+1,π‘šπœ€,+)2 =

=1

1 +π‘š2πœ€2

βˆ‘π‘₯∈Z

(π‘Ž1(π‘₯+1, 𝑦,π‘šπœ€,βˆ’)+π‘šπœ€π‘Ž2(π‘₯+1, 𝑦,π‘šπœ€,+))2+1

1 +π‘š2πœ€2

βˆ‘π‘₯∈Z

(π‘Ž2(π‘₯βˆ’1, 𝑦,π‘šπœ€,+)βˆ’π‘šπœ€π‘Ž1(π‘₯βˆ’1, 𝑦,π‘šπœ€,βˆ’))2 =

=1

1 +π‘š2πœ€2

βˆ‘π‘₯∈Z

(π‘Ž1(π‘₯, 𝑦,π‘šπœ€,βˆ’) +π‘šπœ€π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+))2 +1

1 +π‘š2πœ€2

βˆ‘π‘₯∈Z

(π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+)βˆ’π‘šπœ€π‘Ž1(π‘₯, 𝑦,π‘šπœ€,+))2 =

=βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦,π‘šπœ€,βˆ’)2 + π‘Ž2(π‘₯, 𝑦,π‘šπœ€,+)2

]=

βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦,π‘šπœ€).

Let us now solve the analogue of Problem 9 for π‘šπœ€ = 1. Let us find π‘Ž1(π‘₯, 𝑦) for 𝑦 > |π‘₯|. We need to considerpaths with an odd number of turns only. Denote by 2π‘Ÿ+1 the number of turns in a path. Denote by 𝑅 and 𝐿 thenumber of upwards-right and upwards-left moves respectively. Let π‘₯1, π‘₯2, . . . , π‘₯π‘Ÿ+1 be the number of upwards-rightmoves before the first, the third, . . . , the last turn respectively. Let 𝑦1, 𝑦2, . . . , π‘¦π‘Ÿ+1 be the number of upwards-leftmoves after the first, the third, . . . , the last turn respectively. Then π‘₯π‘˜, π‘¦π‘˜ β‰₯ 1 for 1 ≀ π‘˜ ≀ π‘Ÿ + 1 and

𝑅 = π‘₯1 + Β· Β· Β·+ π‘₯π‘Ÿ+1;

𝐿 = 𝑦1 + Β· Β· Β·+ π‘¦π‘Ÿ+1.

The problem now reduces to a combinatorial one: find the number of positive integer solutions of the resultingequations. For the first equation, this number equals to the number of ways to put π‘Ÿ sticks between 𝑅 coins in arow, that is,

(π‘…βˆ’1π‘Ÿ

). Thus

π‘Ž1(π‘₯, 𝑦,π‘šπœ€) = (1 +π‘š2πœ€2)(1βˆ’π‘¦)/2

βŒŠπ‘¦/2βŒ‹βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ(π‘…βˆ’ 1

π‘Ÿ

)(πΏβˆ’ 1

π‘Ÿ

)(π‘šπœ€)2π‘Ÿ+1.

Since 𝐿+𝑅 = 𝑦 and π‘…βˆ’ 𝐿 = π‘₯, the required formula follows. The proof of the formula for π‘Ž2(π‘₯, 𝑦) for 𝑦 > |π‘₯| isanalogous. The case 𝑦 ≀ |π‘₯| is obvious.15. Answer: for |π‘₯| < 𝑦 we have

limπ‘›β†’βˆž

𝑛 οΏ½οΏ½(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹,π‘š

𝑛,βˆ’

)= (π‘šπ½0(π‘š

βˆšπ‘¦2 βˆ’ π‘₯2), 0);

limπ‘›β†’βˆž

𝑛 οΏ½οΏ½(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹,π‘š

𝑛,+

)= (0,βˆ’π‘š

π‘₯+ π‘¦βˆšπ‘¦2 βˆ’ π‘₯2

𝐽1(π‘šβˆšπ‘¦2 βˆ’ π‘₯2));

for |π‘₯| β‰₯ 𝑦 both limits vanish.

22

Remark. The limit coincides with the retarded fundamental solution of Dirac’s equation, describing motion ofan electron in the plane.

Solution by Ivan Gaidai-Turlov, Timofey Kovalev and Alexey Lvov, participants of Summer conference of

Tournament of towns. Denote π‘₯𝑛 := 2βŒŠπ‘›π‘₯2 βŒ‹, 𝑦𝑛 := 2βŒŠπ‘›π‘¦2 βŒ‹, 𝐴 := π‘₯𝑛+𝑦𝑛2 , 𝐡 := π‘¦π‘›βˆ’π‘₯𝑛

2 .

Lemma 2. limπ‘›β†’βˆž

(1 + (π‘šπ‘› )2)

π‘¦π‘›βˆ’12 = 1.

Proof. We have 1 < (1 + (π‘šπ‘› )2)

π‘¦π‘›βˆ’12 < (1 + (π‘šπ‘› )

2)𝑛𝑦, therefore if 𝑛 β†’ ∞ then (1 + (π‘šπ‘› )2)𝑛𝑦 = 𝑛

√(1 + (π‘šπ‘› )

2)𝑛2𝑦 β†’π‘›βˆšπ‘’π‘š2𝑦 β†’ 1. Hence by the squeeze theorem we have lim

π‘›β†’βˆž(1 + (π‘šπ‘› )

2)π‘¦π‘›βˆ’1

2 = 1.

Lemma 3. limπ‘›β†’βˆž

𝑛(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1 = (βˆ’1)𝑙 Β· (π‘₯+𝑦2

)𝑙( π‘¦βˆ’π‘₯2

)π‘™Β·π‘š2𝑙+1

(𝑙!)2.

Proof. Clearly, 𝑛 Β· (βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1 = (βˆ’1)𝑙 Β· (π΄βˆ’1)(π΄βˆ’2)...(π΄βˆ’π‘™)Β·(π΅βˆ’1)(π΅βˆ’2)...(π΅βˆ’π‘™)(𝑙!)2

Β· π‘š2𝑙+1

𝑛2𝑙 .

Thus limπ‘›β†’βˆž

π΄βˆ’π‘–π‘› = lim

π‘›β†’βˆžπ΄π‘› = lim

π‘›β†’βˆžβŒŠπ‘›π‘₯

2βŒ‹+βŒŠπ‘›π‘¦

2βŒ‹

𝑛 = limπ‘›β†’βˆž

𝑛π‘₯2+𝑛𝑦

2+π‘œ(𝑛)

𝑛 = π‘₯+𝑦2 . Analogously, lim

π‘›β†’βˆžπ΅βˆ’π‘–π‘› = π‘¦βˆ’π‘₯

2 .

Therefore,

limπ‘›β†’βˆž

𝑛 Β· (βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘š

𝑛)2𝑙+1 = lim

π‘›β†’βˆž(βˆ’1)𝑙 Β· (π΄βˆ’1)...(π΄βˆ’π‘™)Β·(π΅βˆ’1)...(π΅βˆ’π‘™)

(𝑙!)2Β· π‘š

2𝑙+1

𝑛2𝑙= (βˆ’1)𝑙 Β·

(π‘₯+𝑦2 )𝑙(π‘¦βˆ’π‘₯

2 )𝑙 Β·π‘š2𝑙+1

(𝑙!)2.

For each 𝑛 we have |(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1| < |(βˆ’1)𝑙 Β· (π‘₯+𝑦2

)𝑙( π‘¦βˆ’π‘₯2

)π‘™Β·π‘š2𝑙+1

(𝑙!)2| because π΄βˆ’π‘–

𝑛 < 𝐴𝑛 < 𝑦+π‘₯

2 andπ΅βˆ’π‘–π‘› < 𝐡

𝑛 < π‘¦βˆ’π‘₯2 .

Lemma 4. The series 𝑛 Β·βˆžβˆ‘π‘™=0

(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1 converges absolutely for all 𝑛.

Proof. We have

𝑛 Β·βˆžβˆ‘π‘™=0

|(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘š

𝑛)2𝑙+1| <

βˆžβˆ‘π‘™=0

|(βˆ’1)𝑙 Β·(π‘₯+𝑦

2 )𝑙(π‘¦βˆ’π‘₯2 )𝑙 Β·π‘š2𝑙+1

(𝑙!)2| =

=

βˆžβˆ‘π‘™=0

π‘š Β·(π‘₯+𝑦

2 )𝑙(π‘¦βˆ’π‘₯2 )𝑙 Β·π‘š2𝑙

(𝑙!)2<

βˆžβˆ‘π‘™=0

π‘š Β·(π‘₯+𝑦

2 )𝑙(π‘¦βˆ’π‘₯2 )𝑙 Β·π‘š2𝑙

(𝑙!)= π‘š Β· 𝑒

𝑦2βˆ’π‘₯2

4π‘š2

.

Hence limπ‘›β†’βˆž

βˆžβˆ‘π‘™=0

(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1 = ( limπ‘›β†’βˆž

βˆžβˆ‘π‘™=0;2|𝑙

𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1)βˆ’ ( limπ‘›β†’βˆž

βˆžβˆ‘π‘™=0;2-𝑙

𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘šπ‘› )

2𝑙+1).

Lemma 5. Suppose ({π‘Ž0(𝑛)}, {π‘Ž1(𝑛)} . . . ) is a sequence of nonnegative sequences such that limπ‘›β†’βˆž

π‘Žπ‘–(𝑛) = 𝑏𝑖 and

π‘Žπ‘–(𝑛) < 𝑏𝑖 for each 𝑖, 𝑛. Then limπ‘›β†’βˆž

βˆžβˆ‘π‘–=0

π‘Žπ‘–(𝑛) =βˆžβˆ‘π‘–=0

𝑏𝑖.

Proof. Denote 𝑏 :=βˆžβˆ‘π‘–=0

𝑏𝑖. Then for each 𝑛 we haveβˆžβˆ‘π‘–=0

π‘Žπ‘–(𝑛) < 𝑏. Take any πœ€ > 0. Take such 𝑁 thatπ‘βˆ‘π‘–=0

𝑏𝑖 > π‘βˆ’ πœ€.

For each 0 ≀ 𝑖 ≀ 𝑁 take 𝑀𝑖 such that βˆ€π‘‘ β‰₯ 𝑀𝑖 : π‘Žπ‘–(𝑑) > 𝑏𝑖 βˆ’ πœ€2𝑖+1 . Then for all 𝑛 > max(𝑀0,𝑀1, . . . ,𝑀𝑁 ) we

haveβˆžβˆ‘π‘–=0

π‘Žπ‘–(𝑛) > π‘βˆ’ πœ€. Therefore, limπ‘›β†’βˆž

βˆžβˆ‘π‘–=0

π‘Žπ‘–(𝑛) = 𝑏.

From Lemmas 5 and 3 it follows that

limπ‘›β†’βˆž

𝑛 Β·βˆžβˆ‘

𝑙=0;2|𝑙

𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘š

𝑛)2𝑙+1 =

βˆžβˆ‘π‘™=0;2|𝑙

(π‘₯+𝑦2 )𝑙(π‘¦βˆ’π‘₯

2 )𝑙 Β·π‘š2𝑙+1

(𝑙!)2.

Therefore, by Problem 14 and Lemma 2 we have

limπ‘›β†’βˆž

𝑛��(π‘₯𝑛, 𝑦𝑛,π‘š

𝑛,βˆ’) = lim

π‘›β†’βˆžπ‘› Β·

βˆžβˆ‘π‘™=0

𝑛(βˆ’1)𝑙𝐢 π‘™π΄βˆ’1𝐢

π‘™π΅βˆ’1(

π‘š

𝑛)2𝑙+1 =

= (

βˆžβˆ‘π‘™=0;2|𝑙

Β·(π‘₯+𝑦

2 )𝑙(π‘¦βˆ’π‘₯2 )𝑙 Β·π‘š2𝑙+1

(𝑙!)2)βˆ’ (

βˆžβˆ‘π‘™=0;2-𝑙

Β·(π‘₯+𝑦

2 )𝑙(π‘¦βˆ’π‘₯2 )𝑙 Β·π‘š2𝑙+1

(𝑙!)2) = π‘š Β· 𝐽0(π‘š

βˆšπ‘¦2 βˆ’ π‘₯2).

23

It can be proven analogously that limπ‘›β†’βˆž

𝑛��(π‘₯𝑛, 𝑦𝑛,π‘šπ‘› ,+) = βˆ’π‘–π‘š π‘₯+π‘¦βˆš

𝑦2βˆ’π‘₯2Β· 𝐽1(π‘š

βˆšπ‘¦2 βˆ’ π‘₯2).

Solution. Denote modified Bessel functions

𝐼0(𝑧) :=βˆžβˆ‘π‘˜=0

(𝑧/2)2π‘˜

(π‘˜!)2and 𝐼1(𝑧) :=

βˆžβˆ‘π‘˜=0

(𝑧/2)2π‘˜+1

π‘˜!(π‘˜ + 1)!.

The answer follows from the following stronger theorem.

Theorem 2. Let 0 < πœ€0 ≀ 1/2, min{𝑦 + π‘₯, 𝑦 βˆ’ π‘₯} β‰₯ πœ€0, 𝑧 := π‘šβˆšπ‘¦2 βˆ’ π‘₯2, 𝑛 β‰₯ 𝑛0 = 𝑛0(𝑧, πœ€0) = 4𝑒3𝑧/πœ€20. Then

π‘Ž1

(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹,π‘š

𝑛,βˆ’

)=

π‘š

𝑛

(𝐽0(𝑧) +𝑂

(log2 𝑛

πœ€0𝑛𝐼0(𝑧)

)),

π‘Ž2

(2βŒŠπ‘›π‘₯2

βŒ‹, 2βŒŠπ‘›π‘¦2

βŒ‹,π‘š

𝑛,+

)= βˆ’π‘š

𝑛

𝑦 + π‘₯βˆšπ‘¦2 βˆ’ π‘₯2

(𝐽1(𝑧) +𝑂

(log2 𝑛

πœ€0𝑛𝐼1(𝑧)

)).

Proof. Denote π‘₯𝑛 = 2βŒŠπ‘›π‘₯2

βŒ‹, 𝑦𝑛 = 2

βŒŠπ‘›π‘¦2

βŒ‹. By Problem 9 we have

π‘Ž1

(π‘₯𝑛, 𝑦𝑛,

π‘š

𝑛,βˆ’

)=

(1 +

π‘š2

𝑛2

) 1βˆ’π‘¦π‘›2 π‘š

𝑛

[𝑦𝑛/2]βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)(π‘šπ‘›

)2π‘Ÿ,

π‘Ž2

(π‘₯𝑛, 𝑦𝑛,

π‘š

𝑛,+

)=

(1 +

π‘š2

𝑛2

) 1βˆ’π‘¦π‘›2 π‘š

𝑛

[𝑦𝑛/2]βˆ‘π‘Ÿ=1

(βˆ’1)π‘Ÿ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ βˆ’ 1

)(π‘šπ‘›

)2π‘Ÿβˆ’1.

Since 1 + π‘š2

𝑛2 ≀ π‘’π‘š2/𝑛2

it follows that(1 + π‘š2

𝑛2

) 1βˆ’π‘¦π‘›2 β‰₯ 𝑒

π‘š2

𝑛2 (1βˆ’π‘¦π‘›

2 ) because 𝑦𝑛 β‰₯ 1. Also 1βˆ’π‘¦π‘›2 β‰₯ βˆ’π‘¦π‘›

2 . Hence,

1 β‰₯(1 +

π‘š2

𝑛2

) 1βˆ’π‘¦π‘›2

β‰₯ π‘’π‘š2

𝑛2 (βˆ’π‘¦π‘›2 ) β‰₯ 1βˆ’ π‘š2𝑦

2𝑛.

Thus(1 + π‘š2

𝑛2

) 1βˆ’π‘¦π‘›2

= 1 +𝑂(π‘š2𝑦𝑛 ).

To find the asymptotic form for π‘Ž1 and π‘Ž2 consider the 𝑇 -th partial sums with 𝑇 = ⌊log π‘›βŒ‹+1 summands. Theinequalities 𝑦 β‰₯ πœ€0 and 𝑛 β‰₯ 4

πœ€20guarantee that the total number of summands βŒŠπ‘¦π‘›2 βŒ‹ β‰₯ 𝑇 because 𝑛𝑦

2 β‰₯βˆšπ‘› β‰₯ log 𝑛+1

for 𝑛 > 0.For π‘Ÿ β‰₯ 𝑇 the ratios of consecutive summands in the expressions for π‘Ž1 and π‘Ž2 equal respectively(π‘š

𝑛

)2 ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2βˆ’ π‘Ÿ)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2βˆ’ π‘Ÿ)

(π‘Ÿ + 1)2<

(π‘šπ‘›

)2Β· 𝑛(𝑦 + π‘₯)

2𝑇· 𝑛(𝑦 βˆ’ π‘₯)

2𝑇=

𝑧2

4𝑇 2,(π‘š

𝑛

)2 ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2βˆ’ π‘Ÿ)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2βˆ’ π‘Ÿ + 1)

(π‘Ÿ βˆ’ 1)π‘Ÿ<

(π‘šπ‘›

)2Β· 𝑛(𝑦 + π‘₯)

2𝑇· 𝑛(𝑦 βˆ’ π‘₯)

2𝑇 βˆ’ 2=

𝑧2

4𝑇 (𝑇 βˆ’ 1).

From the condition 𝑛 > 𝑒3𝑧+1 it follows that 𝑇 = ⌊log π‘›βŒ‹ + 1 β‰₯ 3𝑧 + 1 and 𝑧2

4𝑇 2 < 𝑧2

4𝑇 (π‘‡βˆ’1) < 12 . Therefore, in

both cases the error term (i.e., the sum over π‘Ÿ β‰₯ 𝑇 ) is less then the sum of geometric series with ratio 12 . Hence,

π‘Ž1

(𝑛π‘₯, 𝑛𝑦,

π‘š

𝑛,βˆ’

)=

π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β·

Β·

[π‘‡βˆ’1βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)(π‘šπ‘›

)2π‘Ÿ+𝑂

((𝑦2 βˆ’ π‘₯2)𝑇

(𝑇 !)2Β·(π‘š2

)2𝑇)]

,

π‘Ž2

(𝑛π‘₯, 𝑛𝑦,

π‘š

𝑛,+

)=

π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β·

Β·

[π‘‡βˆ’1βˆ‘π‘Ÿ=1

(βˆ’1)π‘Ÿ((𝑦𝑛 + π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ

)((𝑦𝑛 βˆ’ π‘₯𝑛 βˆ’ 2)/2

π‘Ÿ βˆ’ 1

)(π‘šπ‘›

)2π‘Ÿβˆ’1+𝑂

((𝑦 + π‘₯)𝑇 (𝑦 βˆ’ π‘₯)π‘‡βˆ’1

(𝑇 βˆ’ 1)!𝑇 !Β·(π‘š2

)2π‘‡βˆ’1)]

.

The error term in the latter formulae are estimated as follows. Since 𝑇 ! β‰₯ (𝑇/3)𝑇 and 𝑇 β‰₯√

𝑦2 βˆ’ π‘₯2 3π‘šβˆšπ‘’

2 =3βˆšπ‘’

2 𝑧, it follows that

(𝑦2 βˆ’ π‘₯2)𝑇

(𝑇 !)2Β·(π‘š2

)2𝑇≀ (𝑦2 βˆ’ π‘₯2)𝑇

(𝑇 )2𝑇·(3π‘š

2

)2𝑇

≀ π‘’βˆ’π‘‡ ≀ 1

𝑛.

24

Now transform binomial coefficients in the following way:(π›Όπ‘›βˆ’ 1

π‘Ÿ

)=

(π›Όπ‘›βˆ’ 1) Β· Β· Β· (π›Όπ‘›βˆ’ π‘Ÿ)

π‘Ÿ!=

(𝛼𝑛)π‘Ÿ

π‘Ÿ!

(1βˆ’ 1

𝛼𝑛

)Β· Β· Β·

(1βˆ’ π‘Ÿ

𝛼𝑛

).

If 𝛼 β‰₯ πœ€0 and π‘Ÿ < 𝑇 then π‘Ÿπ›Όπ‘› < 𝑇

𝛼𝑛 <βˆšπ‘›

π‘›πœ€0= 1√

π‘›πœ€0≀ πœ€0

2 Β· 1πœ€0

= 12 . Hence from the inequalities π‘’βˆ’2π‘₯ ≀ 1 βˆ’ π‘₯ < π‘’βˆ’π‘₯

for 0 ≀ π‘₯ ≀ 1/2 it follows that(1βˆ’ 1

𝛼𝑛

)Β· Β· Β·

(1βˆ’ π‘Ÿ

𝛼𝑛

)β‰₯ 𝑒

βˆ’2𝛼𝑛 𝑒

βˆ’4𝛼𝑛 Β· Β· Β· 𝑒

βˆ’2π‘Ÿπ›Όπ‘› = 𝑒

βˆ’π‘Ÿ(π‘Ÿ+1)𝛼𝑛 β‰₯ 𝑒

βˆ’π‘‡2

𝛼𝑛 β‰₯ 1βˆ’ 𝑇 2

𝛼𝑛β‰₯ 1βˆ’ 𝑇 2

πœ€0𝑛.

Therefore,(π›Όπ‘›βˆ’1

π‘Ÿ

)= (𝛼𝑛)π‘Ÿ

π‘Ÿ!

(1 +𝑂

(𝑇 2

πœ€0𝑛

)).

Denote 𝑅 = 𝑇 2

πœ€0𝑛𝐼0(𝑧). Replacing

((𝑦𝑛+π‘₯π‘›βˆ’2)/2

π‘Ÿ

)and

((π‘¦π‘›βˆ’π‘₯π‘›βˆ’2)/2

π‘Ÿ

)in the previous expression for π‘Ž1 we get

π‘Ž1

(π‘₯𝑛, 𝑦𝑛,

π‘š

𝑛,βˆ’

)=

π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β·

[π‘‡βˆ’1βˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ(π‘š2

)2π‘Ÿ (𝑦2 βˆ’ π‘₯2)π‘Ÿ

(π‘Ÿ!)2+𝑂(𝑅)

]=

=π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β·

[ βˆžβˆ‘π‘Ÿ=0

(βˆ’1)π‘Ÿ(π‘š2

)2π‘Ÿ (𝑦2 βˆ’ π‘₯2)π‘Ÿ

(π‘Ÿ!)2+𝑂(𝑅)

]=

π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β· (𝐽0(𝑧) +𝑂(𝑅)).

Here we can replace the sum with 𝑇 summands by an infinite sum because the β€œtail” of alternating series withdecreasing absolute value of the summands can be estimated by the first summand:

βˆžβˆ‘

π‘Ÿ=𝑇

(βˆ’1)π‘Ÿ(π‘š2

)2π‘Ÿ (𝑦2 βˆ’ π‘₯2)π‘Ÿ

(π‘Ÿ!)2

≀ (𝑧/2)2𝑇

(𝑇 !)2≀ 1

𝑛

(the latter inequality has been proved before).We have π‘š2π‘¦πœ€0 ≀ π‘š2 Β· max{𝑦 + π‘₯, 𝑦 βˆ’ π‘₯} Β· min{𝑦 + π‘₯, 𝑦 βˆ’ π‘₯} = π‘š2(𝑦2 βˆ’ π‘₯2) = 𝑧2 ≀ 9𝑧2 ≀ 𝑇 2, thus

π‘š2𝑦𝑛 𝐽0(𝑧) ≀ 𝑇 2

πœ€0𝑛𝐼0(𝑧), hence π‘Ž1

(π‘₯𝑛, 𝑦𝑛,

π‘šπ‘› ,βˆ’

)= π‘š

𝑛 (𝐽0(𝑧) +𝑂(𝑅)) as required.Analogously,

π‘Ž2

(π‘₯𝑛, 𝑦𝑛,

π‘š

𝑛,+

)=

π‘š

𝑛

(1 +𝑂

(π‘š2𝑦

𝑛

))Β· 𝑦 + π‘₯√

𝑦2 βˆ’ π‘₯2Β·

[π‘‡βˆ’1βˆ‘π‘Ÿ=1

(βˆ’1)π‘Ÿ(π‘š2

)2π‘Ÿβˆ’1 (𝑦2 βˆ’ π‘₯2)2π‘Ÿβˆ’1

2

(π‘Ÿ βˆ’ 1)!π‘Ÿ!+𝑂(

𝑇 2

πœ€0𝑛𝐼1(𝑧))

]=

= βˆ’π‘š

𝑛· 𝑦 + π‘₯√

𝑦2 βˆ’ π‘₯2

(𝐽1(𝑧) +𝑂

(𝑇 2

πœ€0𝑛𝐼1(𝑧)

)).

16. Answer : see the following table, where the pair (𝑐, 𝑑)+ in the cell (π‘₯, 𝑦) means that οΏ½οΏ½(π‘₯, 𝑦, 4, 3,+) = 1√3(𝑐, 𝑑),

and (𝑐, 𝑑)βˆ’ means that οΏ½οΏ½(π‘₯, 𝑦, 4, 3,βˆ’) = 1√3(𝑐, 𝑑); an empty cell (π‘₯, 𝑦) means that οΏ½οΏ½(π‘₯, 𝑦, 4, 3,+) = οΏ½οΏ½(π‘₯, 𝑦, 4, 3,βˆ’) =

(0, 0).

3 (0, 0)+(βˆ’1, 0)βˆ’

(0, 1)+(0, 0)βˆ’

(0,βˆ’2)+(0, 0)βˆ’

(0, 2)+(βˆ’1, 0)βˆ’

(0,βˆ’1)+(0, 0)βˆ’

2 (0, 0)+(βˆ’1, 0)βˆ’

(0,βˆ’1)+(1, 0)βˆ’

(0, 1)+(βˆ’1, 0)βˆ’

(0,βˆ’1)+(0, 0)βˆ’

1 (0,βˆ’1)+(0, 0)βˆ’

(0, 1)+(0, 0)βˆ’

(0,βˆ’1)+(0, 0)βˆ’

y x βˆ’3 βˆ’2 βˆ’1 0 1 2 3 4 5

Hint. Note that we are interested only in even π‘₯0, thus 2πœ‹π‘₯0πœ€/πœ† = πœ‹π‘₯02 is a multiple of πœ‹. Analogously to

Problem 1, that means that οΏ½οΏ½(π‘₯, 𝑦, 4, 3,+) is parallel to 𝑂𝑦 and οΏ½οΏ½(π‘₯, 𝑦, 4, 3,βˆ’) is parallel to 𝑂π‘₯ (be careful: this isnot true for an arbitrary πœ€/πœ†).

17. Let us give a more general formula; the solution of the problem is the particular case π‘šπœ€ = 1. Denote

π‘Žβˆ’(π‘₯, 𝑦) := οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€, πœ†/πœ€,Ξ”,βˆ’);

π‘Ž+(π‘₯, 𝑦) := οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€, πœ†/πœ€,Ξ”,+),

25

where the vectors in the right-hand side are defined analogously to οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Ξ”,Β±) and οΏ½οΏ½(π‘₯, 𝑦,π‘šπœ€,Β±). (Bewarethat π‘Žβˆ’(π‘₯, 𝑦) and π‘Ž+(π‘₯, 𝑦) are vector -valued functions; unlike π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦) they are not coordinates of anyvector.) Then we have the Dirac equation (where 𝑖 := π‘…πœ‹/2 is the counterclockwise rotation through 90∘){

π‘Žβˆ’(π‘₯, 𝑦) = 1√1+π‘š2πœ€2

π‘Žβˆ’(π‘₯+ 1, 𝑦 βˆ’ 1) + βˆ’π‘–π‘šπœ€βˆš1+π‘š2πœ€2

π‘Ž+(π‘₯+ 1, 𝑦 βˆ’ 1);

π‘Ž+(π‘₯, 𝑦) = βˆ’π‘–π‘šπœ€βˆš1+π‘š2πœ€2

π‘Žβˆ’(π‘₯βˆ’ 1, 𝑦 βˆ’ 1) + 1√1+π‘š2πœ€2

π‘Ž+(π‘₯βˆ’ 1, 𝑦 βˆ’ 1).

From Problem 14 we know that this equation holds for Ξ” = 1. It can be generalised for any greater Ξ” becauseone can change the order of rotations 𝑅2πœ‹π‘₯0πœ€/πœ† and 𝑖 = π‘…πœ‹/2.

Using this equation, analogously to Problem 14 we getβˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦, πœ†/πœ€,Ξ”) = 1 for 𝑦 β‰₯ 1 by induction over 𝑦.

18. Since Δ𝑃 (π‘₯, 𝑦, πœ†/πœ€,Ξ”,+) = |βˆšΞ”οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Ξ”,+)|2, it suffices to prove that

βˆšΞ”οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Ξ”,+) does not

depend on Ξ” for Ξ” > 𝑦+ |π‘₯|. By definitionβˆšΞ”οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,Ξ”,+) =

π‘₯0=Ξ”βˆ’1βˆ‘π‘₯0=1βˆ’Ξ”π‘₯0 even

βˆ‘π‘ π‘…2πœ‹π‘₯0πœ€/πœ†οΏ½οΏ½(𝑠). But there are no paths

𝑠 starting at (π‘₯0, 0) and ending at (π‘₯, 𝑦) for |π‘₯βˆ’ π‘₯0| > 𝑦, thus increasing Ξ” over 𝑦 + |π‘₯| does not change the sum.

19. Answer: if π‘₯1 and π‘₯2 are both even, then

οΏ½οΏ½(π‘₯1, 𝑦, πœ†/πœ€,+) = 𝑅(π‘₯1βˆ’π‘₯2)2πœ‹πœ€/πœ†οΏ½οΏ½(π‘₯2, 𝑦, πœ†/πœ€,+),

οΏ½οΏ½(π‘₯1, 𝑦, πœ†/πœ€,βˆ’) = 𝑅(π‘₯1βˆ’π‘₯2)2πœ‹πœ€/πœ†οΏ½οΏ½(π‘₯2, 𝑦, πœ†/πœ€,βˆ’).

Solution. From the solution of Problem 18 it follows that οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) =βˆ‘

π‘₯0 even

βˆ‘π‘ π‘…2πœ‹π‘₯0πœ€/πœ†οΏ½οΏ½(𝑠), where the

second sum is over those paths from (π‘₯0, 0) to (π‘₯, 𝑦), which both start and finish with an upwards-right move. Toeach path 𝑠1 from (π‘₯0, 0) to (π‘₯1, 𝑦) assign the unique path 𝑠2 from (π‘₯0 + (π‘₯2 βˆ’ π‘₯1), 0) to (π‘₯2, 𝑦), obtained from 𝑠1by translation by (π‘₯1 βˆ’ π‘₯2, 0) and vice versa. Thus we sum over the same vectors, but for π‘₯ = π‘₯1 they are rotatedthrough 2πœ‹π‘₯0πœ€/πœ† while for π‘₯ = π‘₯2 they are rotated through 2πœ‹(π‘₯0 + (π‘₯2 βˆ’ π‘₯1))πœ€/πœ†, and we are done.

20. Let us give a more general formula; the solution of the problem is the particular case π‘šπœ€ = 1.

Answer. Denote 𝐸 = arccos(cos(2πœ‹πœ€/πœ†)√

1+π‘š2πœ€2

). Then for even π‘₯+ 𝑦 we have:

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’) =(π‘šπœ€ cos(2πœ‹πœ€π‘₯/πœ†) sin((𝑦 βˆ’ 1)𝐸)√

1 +π‘š2πœ€2 sin𝐸,π‘šπœ€ sin(2πœ‹πœ€π‘₯/πœ†) sin((𝑦 βˆ’ 1)𝐸)√

1 +π‘š2πœ€2 sin𝐸

);

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) =(cos(2πœ‹πœ€(π‘₯βˆ’ 2)/πœ†) sin(𝑦𝐸)βˆ’ cos(2πœ‹πœ€π‘₯/πœ†) sin((𝑦 βˆ’ 2)𝐸)

2 cos(2πœ‹πœ€/πœ†) sin𝐸,

βˆ’ sin(2πœ‹πœ€(π‘₯βˆ’ 2)/πœ†) sin(𝑦𝐸) + sin(2πœ‹πœ€π‘₯/πœ†) sin((𝑦 βˆ’ 2)𝐸)

2 cos(2πœ‹πœ€/πœ†) sin𝐸

);

𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’) =π‘š2πœ€2 sin2((𝑦 βˆ’ 1)𝐸)

π‘š2πœ€2 + sin2(2πœ‹πœ€/πœ†);

𝑃 (π‘₯, 𝑦, πœ†/πœ€,+) =π‘š2πœ€2 cos2((𝑦 βˆ’ 1)𝐸) + sin2(2πœ‹πœ€/πœ†)

π‘š2πœ€2 + sin2(2πœ‹πœ€/πœ†);

𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’) + 𝑃 (π‘₯, 𝑦, πœ†/πœ€,+) =1.

Remark. In complex form,

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’) = π‘Ÿπ‘₯π‘šπœ€ sin((𝑦 βˆ’ 1)𝐸)√1 +π‘š2πœ€2 sin𝐸

;

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) = π‘Ÿπ‘₯βˆ’1(sin((𝑦 βˆ’ 1)𝐸) sin(2πœ‹πœ€/πœ†)√

1 +π‘š2πœ€2 sin𝐸+ 𝑖 cos((𝑦 βˆ’ 1)𝐸)

),

where π‘Ÿ = cos(2πœ‹πœ€/πœ†) + 𝑖 sin(2πœ‹πœ€/πœ†). This is the solution of the Dirac equation (see the solution of Problem 17)with the (quasi)periodic initial condition π‘Žβˆ’(π‘₯, 0) = 0 and π‘Ž+(π‘₯, 0) = (βˆ’ sin 2πœ‹π‘₯πœ€

πœ† , cos 2πœ‹π‘₯πœ€πœ† ). The number 𝐸 has

physical meaning of energy.Hint. Formulas for οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) and οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’) can be proven by induction over 𝑦. The base is obvious;

the inductive step follows from Problem 17. Then 𝑃 (π‘₯, 𝑦, πœ†/πœ€,+) and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’) are computed directly.Path to solution. Let us view vectors as complex numbers. Then 𝑅𝛼�� = (cos𝛼 + 𝑖 sin𝛼)π‘Ž. Denote π‘Ÿ =

cos(2πœ‹πœ€/πœ†) + 𝑖 sin(2πœ‹πœ€/πœ†); then π‘Ÿβˆ’1 = cos(2πœ‹πœ€/πœ†) βˆ’ 𝑖 sin(2πœ‹πœ€/πœ†). Denote 𝑓(𝑦) = π‘Ÿβˆ’π‘¦οΏ½οΏ½(𝑦, 𝑦, πœ†/πœ€,+) and 𝑔(𝑦) =π‘Ÿβˆ’π‘¦οΏ½οΏ½(𝑦, 𝑦, πœ†/πœ€,βˆ’). Denote𝑀 = π‘šπœ€βˆš

1+π‘š2πœ€2and π‘ˆ = 1√

1+π‘š2πœ€2. By Problem 19 for even π‘₯+𝑦 we have οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) =

π‘Ÿπ‘₯𝑓(𝑦) and οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’) = π‘Ÿπ‘₯𝑔(𝑦). By Problem 17 we have π‘Ÿπ‘₯𝑓(𝑦) = π‘ˆπ‘Ÿπ‘₯βˆ’1𝑓(𝑦 βˆ’ 1) βˆ’ π‘–π‘€π‘Ÿπ‘₯βˆ’1𝑔(𝑦 βˆ’ 1) and

26

π‘Ÿπ‘₯𝑔(𝑦) = π‘ˆπ‘Ÿπ‘₯+1𝑔(𝑦 βˆ’ 1) βˆ’ π‘–π‘€π‘Ÿπ‘₯+1𝑓(𝑦 βˆ’ 1). Since π‘Ÿ = 0, we can divide each side by π‘Ÿπ‘₯ and obtain the followingsystem: ⎧βŽͺβŽͺβŽͺ⎨βŽͺβŽͺβŽͺ⎩

𝑓(𝑦) = π‘ˆπ‘Ÿβˆ’1𝑓(𝑦 βˆ’ 1)βˆ’ π‘–π‘€π‘Ÿβˆ’1𝑔(𝑦 βˆ’ 1) (1)

𝑔(𝑦) = π‘ˆπ‘Ÿπ‘”(𝑦 βˆ’ 1)βˆ’ π‘–π‘€π‘Ÿπ‘“(𝑦 βˆ’ 1) (2)

𝑓(1) = π‘–π‘Ÿβˆ’1 (3)

𝑔(1) = 0 (4)

There are multiple ways of solving this system (we have thought about generating functions and about diagonalisingthe β€˜transfer matrix’) but the one which is probably most accessible (to math high-school students in Russia) istransforming this system into a linear recurrence relation of degree 2. From (1) we have 𝑔(𝑦 βˆ’ 1) = 1

𝑀 π‘–π‘Ÿπ‘“(𝑦) βˆ’π‘ˆπ‘€ 𝑖𝑓(𝑦 βˆ’ 1). From this and (2) we have 𝑔(𝑦) = π‘ˆ

𝑀 π‘–π‘Ÿ2𝑓(𝑦) βˆ’ 1𝑀 π‘–π‘Ÿπ‘“(𝑦 βˆ’ 1). From this and (1) we have 𝑓(𝑦) =

π‘ˆ(π‘Ÿ + π‘Ÿβˆ’1)𝑓(𝑦 βˆ’ 1)βˆ’ 𝑓(𝑦 βˆ’ 2).The solution of this recurrence equation is 𝑓(𝑛) = 𝑐1𝑑

𝑛+ + 𝑐2𝑑

π‘›βˆ’, where

𝑑+ =π‘ˆ(π‘Ÿ + π‘Ÿβˆ’1) +

βˆšπ‘ˆ2(π‘Ÿ + π‘Ÿβˆ’1)2 βˆ’ 4

2= π‘ˆ cos(2πœ‹πœ€/πœ†) + 𝑖

βˆšπ‘ˆ2 sin2(2πœ‹πœ€/πœ†) +𝑀2

π‘‘βˆ’ =π‘ˆ(π‘Ÿ + π‘Ÿβˆ’1)βˆ’

βˆšπ‘ˆ2(π‘Ÿ + π‘Ÿβˆ’1)2 βˆ’ 4

2= π‘ˆ cos(2πœ‹πœ€/πœ†)βˆ’ 𝑖

βˆšπ‘ˆ2 sin2(2πœ‹πœ€/πœ†) +𝑀2

are the roots of the equation 𝑑2βˆ’π‘ˆ(π‘Ÿ+π‘Ÿβˆ’1)𝑑+1 = 0, and 𝑐1 and 𝑐2 are constants that are found using (3) and (4).Since we are dealing with complex numbers, the notation

βˆšπ‘ˆ2(π‘Ÿ + π‘Ÿβˆ’1)2 βˆ’ 4 is ambiguous. But we can choose

any of the two roots;√

π‘ˆ2(π‘Ÿ + π‘Ÿβˆ’1)2 βˆ’ 4 denotes 2π‘–βˆš

π‘ˆ2 sin2(2πœ‹πœ€/πœ†) +𝑀2. It follows that

𝑐1 = π‘–π‘Ÿβˆ’1 βˆ’ π‘ˆπ‘‘βˆ’π‘‘+ βˆ’ π‘‘βˆ’

and 𝑐2 = π‘–π‘Ÿβˆ’1 βˆ’ π‘ˆπ‘‘+π‘‘βˆ’ βˆ’ 𝑑+

.

Note that 𝐸 = arccos(π‘ˆ cos(2πœ‹πœ€/πœ†)). We have 𝑑+ = cos𝐸 + 𝑖 sin𝐸 and π‘‘βˆ’ = cos𝐸 βˆ’ 𝑖 sin𝐸. Thus we have

οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) = π‘Ÿπ‘₯𝑓(𝑦) = π‘–π‘Ÿπ‘₯(π‘Ÿβˆ’1βˆ’π‘ˆπ‘‘βˆ’)𝑑𝑦+βˆ’(π‘Ÿβˆ’1βˆ’π‘ˆπ‘‘+)π‘‘π‘¦βˆ’

𝑑+βˆ’π‘‘βˆ’= π‘–π‘Ÿπ‘₯βˆ’1 sin(𝑦𝐸)

sin𝐸 βˆ’ π‘–π‘ˆπ‘Ÿπ‘₯ sin((π‘¦βˆ’1)𝐸)sin𝐸 .

Now we compute 𝑃 (π‘₯, 𝑦, πœ†/πœ€,+) = οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+) Β· οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,+). It follows from formulas above that π‘Ÿ = π‘Ÿβˆ’1,𝑑+ = π‘‘βˆ’, π‘‘βˆ’ = 𝑑+, 𝑐1 = 𝑖 π‘Ÿβˆ’π‘ˆπ‘‘+

𝑑+βˆ’π‘‘βˆ’and 𝑐2 = 𝑖 π‘Ÿβˆ’π‘ˆπ‘‘βˆ’

π‘‘βˆ’βˆ’π‘‘+. From this (with some simplification) we finally have

𝑃 (π‘₯, 𝑦, πœ†/πœ€,+) =(𝑑+βˆ’π‘‘βˆ’)2βˆ’π‘€2(π‘‘π‘¦βˆ’1

+ βˆ’π‘‘π‘¦βˆ’1βˆ’ )2

(𝑑+βˆ’π‘‘βˆ’)2= 1βˆ’ 𝑀2 sin2((π‘¦βˆ’1)𝐸)

sin2 𝐸.

Analogously we get οΏ½οΏ½(π‘₯, 𝑦, πœ†/πœ€,βˆ’) = π‘Ÿπ‘₯𝑀 sin((π‘¦βˆ’1)𝐸)sin𝐸 and 𝑃 (π‘₯, 𝑦, πœ†/πœ€,βˆ’) = 𝑀2 sin2((π‘¦βˆ’1)𝐸)

sin2 𝐸.

21. Solution. Let (0, 0), (1, 1), (π‘₯2, 2), Β· Β· Β· , (π‘₯𝑦, 𝑦) be the squares passed by the path 𝑠. To define the vectorοΏ½οΏ½(𝑠,π‘š(π‘₯)πœ€), start with the vector (0, 1). While the checker moves straightly, the vector is not changed, but if thechecker changes the direction after the 𝑛-th move, the vector is rotated through 90∘ clockwise and multiplied byπ‘š(π‘₯𝑛)πœ€. In addition, at the very end the vector is divided by

βˆπ‘¦βˆ’1𝑛=1

√1 +π‘š2(π‘₯𝑛)πœ€2, where 𝑦 is the total number

of moves. The final position of the vector is what we denote by οΏ½οΏ½(𝑠,π‘š(π‘₯)πœ€). The numbers π‘Ž1(π‘₯, 𝑦,π‘š(π‘₯)πœ€) andπ‘Ž2(π‘₯, 𝑦,π‘š(π‘₯)πœ€) are defined analogously to π‘Ž1(π‘₯, 𝑦) and π‘Ž2(π‘₯, 𝑦), only οΏ½οΏ½(𝑠) is replaced by οΏ½οΏ½(𝑠,π‘š(π‘₯)πœ€).

The formulas analogous to the ones from Problem 4 are the following:

π‘Ž1(π‘₯0, 𝑦,π‘š(π‘₯)πœ€) =1√

1 +π‘š2(π‘₯0 + 1)πœ€2(π‘Ž1(π‘₯0 + 1, 𝑦,π‘š(π‘₯)πœ€) +π‘š(π‘₯0 + 1)πœ€ Β· π‘Ž2(π‘₯0 + 1, 𝑦,π‘š(π‘₯)πœ€)),

π‘Ž2(π‘₯0, 𝑦,π‘š(π‘₯)πœ€) =1√

1 +π‘š2(π‘₯0 βˆ’ 1)πœ€2(π‘Ž2(π‘₯0 βˆ’ 1, 𝑦,π‘š(π‘₯)πœ€)βˆ’π‘š(π‘₯0 βˆ’ 1)πœ€ Β· π‘Ž1(π‘₯0 βˆ’ 1, 𝑦,π‘š(π‘₯)πœ€)).

22. Answer : For each π‘₯, 𝑦 we have𝑃 (π‘₯, 𝑦,π‘š0(π‘₯), πœ†/πœ€,+) = 1;𝑃 (π‘₯, 𝑦,π‘š0(π‘₯), πœ†/πœ€,βˆ’) = 0.

𝑃 (π‘₯, 𝑦,π‘š1(π‘₯), πœ†/πœ€,+) =

{1

1.04 = 2526 , for 0 < π‘₯ < 𝑦;

1, otherwise;

𝑃 (π‘₯, 𝑦,π‘š1(π‘₯), πœ†/πœ€,βˆ’) =

{0.041.04 = 1

26 , for βˆ’ 𝑦 < π‘₯ < 0;

0, otherwise;

Solution. Analogously to Problem 13 we have 𝑃 (π‘₯, 𝑦,π‘š0(π‘₯), πœ†/πœ€,+) = 1, 𝑃 (π‘₯, 𝑦,π‘š0(π‘₯), πœ†/πœ€,βˆ’) = 0 for eachπ‘₯, 𝑦.

27

The condition οΏ½οΏ½(𝑠,π‘š1(π‘₯)) = 0 means that the checker, if at all changed the direction, turned in one ofthe squares with the coordinates (0, 2), (0, 4), Β· Β· Β· , (0, 2 Β·

[𝑦2

]). Thus the number of turns in any path 𝑠 with

οΏ½οΏ½(𝑠,π‘š1(π‘₯)) = 0 is not greater than 1.To compute 𝑃 (π‘₯, 𝑦,π‘š1(π‘₯), πœ†/πœ€,+) we need to consider only the paths ending with an upwards-right move –

i.e. without turns. If the path 𝑠 does not contain any of the squares with coordinates (0, 1), (0, 2), . . . , (0, 𝑦 βˆ’ 1)(amounting to π‘₯ ≀ 0 or π‘₯ β‰₯ 𝑦) then 𝑃 (π‘₯, 𝑦,π‘š1(π‘₯), πœ†/πœ€,+) = 1. Otherwise, the length of the vector οΏ½οΏ½(𝑠,π‘š(π‘₯))equals 1√

1+0.22.

To compute 𝑃 (π‘₯, 𝑦,π‘š1(π‘₯), πœ†/πœ€,βˆ’), consider paths having exactly one turn in the square with zero π‘₯-coordinate.If there is such a path 𝑠 (i.e. βˆ’π‘¦ < π‘₯ < 0), then the length of the vector οΏ½οΏ½(𝑠,π‘š(π‘₯)) equals 0.2√

1+0.22. Otherwise, it

equals zero.

23. Solution. If οΏ½οΏ½(𝑠,π‘š2(π‘₯)) = 0 then the checker changed the direction only in the squares with the π‘₯-coordinateequal to 1 or 𝐿. To compute 𝑃 (0, 2𝐿,π‘š2(π‘₯), 16,βˆ’) we need to consider only the paths with an odd number ofturns. Since for 𝐿 > 1 we have 𝑦 = 2𝐿 < 3(πΏβˆ’ 1) (for 𝐿 = 1 the function 𝑓(𝐿) is undefined), it follows that thenumber of turns must be exactly 1. Therefore, either the path starts in the square (0, 0) and changes the directionin the square (𝐿,𝐿), or the path starts in the square (2 βˆ’ 2𝐿, 0) and changes the direction in (1, 2𝐿 βˆ’ 1). The

vectors for these two paths are related by the rotation through the angle πœ‹ βˆ’ πœ‹(πΏβˆ’1)4 and division by 1.04. Since

the vector for the latter path has square 126 and the one for the former one has square 1

26Β·1.042 , by the law of cosineswe get

𝑃 (0, 2𝐿,π‘š2(π‘₯), 16,βˆ’) =1

26

(1 +

1

1.042βˆ’ 2

1.04cos

πœ‹(πΏβˆ’ 1)

4

)β‰ˆ 0.074

(1βˆ’ cos

πœ‹(πΏβˆ’ 1)

4

).

The graph of this function is shown in the following figure (kindly provided by Punnawith Thuwajit, a participantof Summer Conference of Tournament of Towns).

24. Answer :

𝑃 (πœ†, 𝐿,βˆ’) =0.1664

1.1664 + cot2 2πœ‹(πΏβˆ’1)πœ†

; max𝐿

𝑃 (πœ†, 𝐿,βˆ’) =0.1664

1.1664β‰ˆ 0.14;

𝑃 (πœ†, 𝐿,+) =1

1 + 0.1664 sin2 2πœ‹(πΏβˆ’1)πœ†

; 𝑃 (πœ†, 𝐿,βˆ’) + 𝑃 (πœ†, 𝐿,+) = 1.

Hint. Denote π‘š = 0.2 and πœ‘ = 2πœ‹πœ† (πΏβˆ’ 1). For 𝑦 > 2πΏβˆ’ 2 we have

𝑃 (0, 𝑦,π‘š2(π‘₯), πœ†,βˆ’) =

=

𝑒 2πœ‹π‘–

πœ†(2βˆ’π‘¦) π‘šβˆš

1 +π‘š2

βŽ›βŽβˆ’1 +1

1 +π‘š2𝑒2π‘–πœ‘ +

π‘š2

(1 +π‘š2)2(𝑒2π‘–πœ‘)2 + Β· Β· Β·+ 1

π‘š2

(π‘š2

1 +π‘š2𝑒2π‘–πœ‘

)[π‘¦βˆ’2

2(πΏβˆ’1)

]2(πΏβˆ’1)

⎞⎠2

,

Then, since cos 2πœ‘ = 1βˆ’ 2 sin2 πœ‘, we have

𝑃 (πœ†, 𝐿,βˆ’) =π‘š2

1 +π‘š2Β·

1βˆ’ 1

1 +π‘š2𝑒2π‘–πœ‘ Β·

βˆžβˆ‘π‘˜=0

(π‘š2

1 +π‘š2𝑒2π‘–πœ‘

)π‘˜2

=

=π‘š2

1 +π‘š2Β·

1βˆ’ 1

1 +π‘š2𝑒2π‘–πœ‘

1

1βˆ’ π‘š2

1+π‘š2 𝑒2π‘–πœ‘

2

=π‘š2

1 +π‘š2Β· |1βˆ’ 𝑒2π‘–πœ‘|2

|1βˆ’ π‘š2

1+π‘š2 𝑒2π‘–πœ‘|2=

=π‘š2(1 +π‘š2)(2βˆ’ 2 cos(2πœ‘))

(1 +π‘š2)2 +π‘š4 βˆ’ 2π‘š2(1 +π‘š2) cos(2πœ‘)=

4π‘š2(1 +π‘š2)

(1 + 2π‘š2)2 + cot2 πœ‘=

0.1664

1.1664 + cot2 2πœ‹πœ† (πΏβˆ’ 1)

.

28

Therefore, max𝐿 𝑃 (πœ†, 𝐿,βˆ’) = 0.16641.1664 β‰ˆ 0.14. Analogously,

𝑃 (πœ†, 𝐿,+) =

1

1 +π‘š2Β·

βˆžβˆ‘π‘˜=0

(π‘š2

1 +π‘š2𝑒2π‘–πœ‘

)π‘˜2

=

1

1 +π‘š2Β· 1

1βˆ’ π‘š2

1+π‘š2 𝑒2π‘–πœ‘

2

=

=1

(1 +π‘š2)2 +π‘š4 βˆ’ 2(1 +π‘š2)π‘š2 cos(2πœ‘)=

1

1 + 4π‘š2(1 +π‘š2) sin2 πœ‘=

1

1 + 0.1664 sin2 2πœ‹πœ† (πΏβˆ’ 1)

.

Note that

1βˆ’ 1

1 + 0.1664 sin2 2πœ‹πœ† (πΏβˆ’ 1)

=0.1664

1sin2 2πœ‹

πœ†(πΏβˆ’1)

+ 0.1664=

0.1664

1.1664 + cot2 2πœ‹πœ† (πΏβˆ’ 1)

.

Hence, 𝑃 (πœ†, 𝐿,βˆ’) + 𝑃 (πœ†, 𝐿,+) = 1.

25. Answer :(𝑛2 βˆ’ 1)2

(𝑛2 + 1)2 + 4𝑛2 cot2 2πœ‹πΏπ‘›πœ†

and π‘š =

(0.3328 + 0.864

√1.04

)Β· πœ‹

πœ†,

where 𝑛 =√1 +π‘šπœ†/πœ‹ (the refractive index ; notice that the relation between π‘š and 𝑛 has changed when we

passed to reflection inside the glass). For the latter π‘š the limit equals

0.1664

1.1664 + cot22πœ‹πΏ(1.08+0.08

√1.04)

πœ†

.

Remark. The first formula in the answer is well-known and confirmed by experiment. A remarkable detailedpopular science discussion of Problems 23–25 can be found in [Feynman], and a derivation of the equations on aphysical level of rigor β€” in [Landafshitz, Chapter X, Β§86, Problem 4].

Path to solution. Consider Dirac’s equation from the solution of Problem 17. Search for a solution of the form

(π‘Ž+(π‘₯, 𝑦), π‘Žβˆ’(π‘₯, 𝑦)) = (π‘Ž+(π‘₯), π‘Žβˆ’(π‘₯))π‘’βˆ’2πœ‹π‘–πœ€π‘¦/πœ†.

Use Problem 20 to guess π‘ŽΒ±(π‘₯) separately for π‘₯ ≀ 0, 0 ≀ π‘₯ ≀ 𝐿/πœ€, and π‘₯ β‰₯ 𝐿/πœ€. Assume for simplicity that 1/πœ€is an odd integer. Abbreviate π‘š3(π‘₯) =: π‘š(π‘₯).

First consider the half-plane π‘₯ ≀ 0 (to the left from glass). We need to specify what is meant by a solutionof Dirac’s equation in the half-plane. Analogously to Problem 17, by Dirac’s equation in a black square (π‘₯, 𝑦) wemean the system βŽ§βŽ¨βŽ©π‘Žβˆ’(π‘₯βˆ’ 1, 𝑦 + 1) = 1√

1+π‘š(π‘₯)2πœ€2π‘Žβˆ’(π‘₯, 𝑦) +

βˆ’π‘–π‘š(π‘₯)πœ€βˆš1+π‘š(π‘₯)2πœ€2

π‘Ž+(π‘₯, 𝑦);

π‘Ž+(π‘₯+ 1, 𝑦 + 1) = βˆ’π‘–π‘š(π‘₯)πœ€βˆš1+π‘š(π‘₯)2πœ€2

π‘Žβˆ’(π‘₯, 𝑦) +1√

1+π‘š(π‘₯)2πœ€2π‘Ž+(π‘₯, 𝑦).

A solution of Dirac’s equation in the half-plane π‘₯ ≀ 0 is a pair (π‘Ž+(π‘₯, 𝑦), π‘Žβˆ’(π‘₯, 𝑦)) satisfying the equation in eachblack square in the half-plane. In particular, π‘Žβˆ’(π‘₯, 𝑦) is defined for π‘₯ ≀ 0, whereas π‘Ž+(π‘₯, 𝑦) is defined for π‘₯ ≀ 1.Search for a solution of the form (π‘Ž+(π‘₯), π‘Žβˆ’(π‘₯)) = (𝑖 𝑒2πœ‹π‘–π‘₯πœ€/πœ†, π‘Ÿ(πœ€) π‘’βˆ’2πœ‹π‘–π‘₯πœ€/πœ†) for some π‘Ÿ(πœ€) ∈ C to be determinedlater, where the formulae for π‘Ž+(π‘₯) and π‘Žβˆ’(π‘₯) hold for π‘₯ ≀ 1 and π‘₯ ≀ 0 respectively. (Physically, π‘Ž+(π‘₯, 𝑦) is theincident wave, π‘Žβˆ’(π‘₯, 𝑦) is the reflected wave, and | limπœ€β†’0 π‘Ÿ(πœ€)|2 is the reflection probability.)

For the half-plane π‘₯ β‰₯ 𝐿/πœ€ take (π‘Ž+(π‘₯), π‘Žβˆ’(π‘₯)) = (𝑑(πœ€) 𝑒2πœ‹π‘–π‘₯πœ€/πœ†, 0) for some 𝑑(πœ€) ∈ C to be determined later,where the formulae for π‘Ž+(π‘₯) and π‘Žβˆ’(π‘₯) hold for π‘₯ β‰₯ 𝐿/πœ€+ 1 and π‘₯ β‰₯ 𝐿/πœ€ respectively. (Physically, π‘Ž+(π‘₯) is thetransmitted wave, and | limπœ€β†’0 𝑑(πœ€)|2 is the transmission probability.)

For 0 ≀ π‘₯ ≀ 𝐿/πœ€ search for a solution of the form

π‘Ž+(π‘₯) = π‘Ž(πœ€) 𝑒2πœ‹π‘–π‘₯πœ€/πœ†β€²(πœ€) + 𝑏(πœ€) π‘’βˆ’2πœ‹π‘–π‘₯πœ€/πœ†β€²(πœ€),

π‘Žβˆ’(π‘₯) = 𝑐(πœ€) 𝑒2πœ‹π‘–π‘₯πœ€/πœ†β€²(πœ€) + 𝑑(πœ€) π‘’βˆ’2πœ‹π‘–π‘₯πœ€/πœ†β€²(πœ€).

for some coefficients π‘Ž, 𝑏, 𝑐, 𝑑 ∈ C and πœ†β€² > 0 depending on πœ€ (we have a combination of refracted and reflectedwaves). In what follows π‘Ž, 𝑏, 𝑐, 𝑑, πœ†β€², π‘Ÿ, 𝑑 denote the limits of the above numbers as πœ€ β†’ 0.

The limit πœ†β€² (wavelength of refracted wave) is determined by Dirac’s equation as follows. Substituting theexpressions for π‘ŽΒ±(π‘₯) into Dirac’s equation, canceling common factors, and taking the limit πœ€ β†’ 0 we get

π‘šπ‘Ž =

(π‘šβˆ’ 2πœ‹

πœ†β€² +2πœ‹

πœ†

)𝑐,

π‘šπ‘ =

(π‘š+

2πœ‹

πœ†β€² +2πœ‹

πœ†

)π‘Ž.

29

We also get an analogous equation with 𝑏 and 𝑑 instead of 𝑐 and π‘Ž. This implies that πœ†β€² = πœ†/𝑛, where 𝑛 =√1 +π‘šπœ†/πœ‹ (refractive index ).Sewing together the solutions (namely, π‘Ž+(π‘₯) at π‘₯ = 1 and π‘₯ = 𝐿/πœ€+1, whereas π‘Žβˆ’(π‘₯) at π‘₯ = 0 and π‘₯ = 𝐿/πœ€)

gives a system of linear equations in π‘Ž, 𝑏, 𝑐, 𝑑, π‘Ÿ, 𝑑:

π‘Žβˆ’(0) = π‘Ÿ(πœ€) = 𝑐(πœ€) + 𝑑(πœ€), π‘Žβˆ’(𝐿/πœ€) = 0 = 𝑐(πœ€) 𝑒2πœ‹π‘–πΏ/πœ†β€²+ 𝑑(πœ€) π‘’βˆ’2πœ‹π‘–πΏ/πœ†β€²

,

π‘Ž+(1) = 𝑖𝑒2πœ‹π‘–πΏ/πœ† = π‘Ž(πœ€) 𝑒2πœ‹π‘–πΏ/πœ†β€²+ 𝑏(πœ€) π‘’βˆ’2πœ‹π‘–πΏ/πœ†β€²

, π‘Ž+(𝐿/πœ€+ 1) = 𝑑(πœ€) 𝑒2πœ‹π‘–(𝐿+πœ€)/πœ† = π‘Ž(πœ€) 𝑒2πœ‹π‘–(𝐿+πœ€)/πœ†β€²+ 𝑏(πœ€) π‘’βˆ’2πœ‹π‘–(𝐿+πœ€)/πœ†β€²

.

Passing to the limit πœ€ β†’ 0 and solving the resulting system of linear equations we find π‘Ž, 𝑏, 𝑐, 𝑑, π‘Ÿ, 𝑑 Surely, one stillneeds to check that even before passing to the limit, the expressions for π‘ŽΒ±(π‘₯) determined by our equations indeedsatisfy Dirac’s equation in each black square. We omit that.

It remains to prove that (π‘Ž+(π‘₯), π‘Žβˆ’(π‘₯)) = lim𝑦→+βˆžπ‘¦+π‘₯ even

𝑅2πœ‹π‘¦πœ€/πœ†οΏ½οΏ½(π‘₯,𝑦,π‘š3(π‘₯)πœ€,πœ†πœ€ ,βˆ’); then the number |π‘Ÿ|2 is the required

limit. We do not know any elementary proof of that assertion and thus omit the details.Informally, the idea is to show that the limit does not depend on the initial values π‘Ž+(π‘₯, 0) for π‘₯ > 0 and

π‘Žβˆ’(π‘₯, 0) for π‘₯ < 𝐿/πœ€; then one can replace these initial values by the solution we have found above and get theresult. Independence on π‘Ž+(π‘₯, 0) for π‘₯ > 𝐿/πœ€ and π‘Žβˆ’(π‘₯, 0) for π‘₯ < 0 is obvious. To prove independence onπ‘ŽΒ±(π‘₯, 0) for 0 ≀ π‘₯ ≀ 𝐿/πœ€, assume that π‘ŽΒ±(π‘₯, 0) vanishes outside the segment 0 ≀ π‘₯ ≀ 𝐿/πœ€ and consider the lineartransformation

π‘ˆ : (π‘Žβˆ’(0, 0), π‘Žβˆ’(2, 0), . . . , π‘Žβˆ’(𝐿/πœ€βˆ’ 1, 0), π‘Ž+(0, 0), . . . , π‘Ž+(𝐿/πœ€βˆ’ 1, 0))

↦→ (π‘Žβˆ’(0, 2), π‘Žβˆ’(2, 2), . . . , π‘Žβˆ’(𝐿/πœ€βˆ’ 1, 2), π‘Ž+(0, 2), . . . , π‘Ž+(𝐿/πœ€βˆ’ 1, 2)).

Let us show that the absolute values of all the eigenvalues of π‘ˆ are less than 1. Consider an eigenvector 𝑣 withthe coordinates 𝑣1 = π‘Žβˆ’(0, 0), . . . , 𝑣𝐿/πœ€+1 = π‘Ž+(𝐿/πœ€ βˆ’ 1, 0). Let π‘£π‘˜ be the leftmost nonzero coordinate. If π‘˜ > 1then π‘ˆπ‘£ has nonzero (π‘˜ βˆ’ 1)-th coordinate; but π‘ˆπ‘£ is proportional to 𝑣, a contradiction. Thus π‘˜ = 1. Henceπ‘Žβˆ’(βˆ’2, 2) = 0 by Dirac’s equation. Then by the probability conservation

‖𝑣‖ = |π‘Žβˆ’(0, 0)|2 + |π‘Žβˆ’(2, 0)|2 + Β· Β· Β·+ |π‘Ž+(𝐿/πœ€βˆ’ 1, 0)|2

= |π‘Žβˆ’(βˆ’2, 2)|2 + |π‘Žβˆ’(0, 2)|2 + Β· Β· Β·+ |π‘Ž+(𝐿/πœ€βˆ’ 1, 2)|2

> |π‘Žβˆ’(0, 2)|2 + Β· Β· Β·+ |π‘Ž+(𝐿/πœ€βˆ’ 1, 2)|2

= β€–π‘ˆπ‘£β€–.

Thus the absolute value of the eigenvalue 𝑣 is strictly less than 1. Thus β€–π‘ˆπ‘¦β€– β†’ 0 as 𝑦 β†’ ∞, which proves thatthe values π‘ŽΒ±(π‘₯, 0) for 0 ≀ π‘₯ ≀ 𝐿 do not affect the limit.

26. Answer : In the table, the vector in the cell (π‘₯, 𝑦) is οΏ½οΏ½(π‘₯, 𝑦, 𝑒,+), and an empty cell (π‘₯, 𝑦) means thatοΏ½οΏ½(π‘₯, 𝑦, 𝑒,+) = (0, 0).

4(0, 1

2√2

)(0, 0)

(0, 1

2√2

)3

(0,βˆ’1

2

) (0, 12

)2

(0,βˆ’ 1√

2

)1 (0,βˆ’1)

y x βˆ’2 βˆ’1 0 1 2 3 4

In the following table the number in the cell (π‘₯, 𝑦) is 𝑃 (π‘₯, 𝑦, 𝑒,+), and an empty cell (π‘₯, 𝑦) means that𝑃 (π‘₯, 𝑦, 𝑒,+) = 0.

4 1/8 0 1/8

3 1/4 1/4

2 1/2

1 1

y x βˆ’2 βˆ’1 0 1 2 3 4

(The vectors οΏ½οΏ½(π‘₯, 𝑦, 𝑒,+) can be easily computed using Problem 31.)For any positive 𝑦 we have

βˆ‘π‘₯∈Z

(𝑃 (π‘₯, 𝑦, 𝑒,+)+𝑃 (π‘₯, 𝑦, 𝑒,βˆ’)) = 1. (This is one of the statements of Problem 31.)

27. Answer (kindly provided by Gleb Minaev and Ivan Russkikh, participants of Summer Conference of Tourna-ment of Towns):

30

28. Solution. Suppose that 𝑒′ is the result of changing the values of 𝑒 in all the vertices of a black square (π‘Ž, 𝑏).Notice that if a path 𝑠 neither starts nor ends in the square (π‘Ž, 𝑏), then it either passes through two vertices witha changed value of 𝑒 or does not pass through such vertices at all. Therefore, οΏ½οΏ½(𝑠, 𝑒) = οΏ½οΏ½(𝑠, 𝑒′). The equation𝑃 (π‘₯, 𝑦, 𝑒′,+) = 𝑃 (π‘₯, 𝑦, 𝑒,+) for any (π‘₯, 𝑦) = (π‘Ž, 𝑏) follows immediately.

The only case left to prove is (π‘Ž, 𝑏) = (π‘₯, 𝑦) or (π‘Ž, 𝑏) = (0, 0). For each path 𝑠 starting or ending in (π‘Ž, 𝑏)we have οΏ½οΏ½(𝑠, 𝑒′) = βˆ’οΏ½οΏ½(𝑠, 𝑒) because 𝑠 passes through exactly one vertex with a changed value of 𝑒. Therefore,𝑃 (π‘Ž, 𝑏, 𝑒′,+) = 𝑃 (π‘Ž, 𝑏, 𝑒,+).29. Solution. The top-right vertex of the square (π‘₯, 𝑦) is called vertex (π‘₯, 𝑦). Denote by 𝑒(π‘₯, 𝑦) the value of 𝑒 atthe vertex (π‘₯, 𝑦). Reversing a black square (π‘₯, 𝑦) means changing the values of 𝑒 at all the four vertices of thesquare.

Notice that reversing a black square does not change the sign of any white squares, because the number ofchanged vertices in each white square is even. If all values of a field in a rectangle are positive, then all the whitesquares inside the rectangle are positive. Therefore, it is impossible to make 𝑒 identically +1 by reversing blacksquares, if initially there is a negative white square.

Suppose that there are no negative white squares in a rectangle 𝑀 ×𝑁 , where 𝑁 is the height of the rectangle.Let the vertex (0, 0) be at the bottom-left of the rectangle. Let us prove that 𝑒 can be made identically +1 throughreversal of black squares by induction over 𝑁 .

Base: 𝑁 = 0. Let us describe the algorithm for making 𝑒 identically +1 in the rectangle 𝑀 Γ— 0 (i.e in vertices(0, 0), (1, 0), . . . , (𝑀, 0)).

Note that for all π‘˜ we can reverse either square (π‘˜+ 1, 0) or square (π‘˜+ 1, 1), thereby we can change the signsof the values of 𝑒 in vertices (π‘˜, 0) and (π‘˜ + 1, 0). Denote this operation reversing π‘˜.

The algorithm is as follows: Find the minimum 0 ≀ π‘˜ ≀ 𝑀 such that 𝑒(π‘˜, 0) = βˆ’1, and reverse π‘˜. If after thatthere still exists 0 ≀ π‘˜ ≀ 𝑀 such that 𝑒(π‘˜, 0) = βˆ’1, than the minimum such π‘˜ is at least 1 more than before usingthe operation. Therefore, after at most 𝑀 + 1 operations we make 𝑒 to be identically +1 in the rectangle 𝑀 Γ— 0.

Step: Suppose that the statement is true for 𝑁 and prove it for 𝑁 + 1 by giving an explicit algorithm.First make 𝑒 identically +1 in the lower rectangle 𝑀 Γ— 𝑁 by applying the algorithm for 𝑁 . Then for each

white square (π‘˜,𝑁 + 1) in the rectangle reverse the square (π‘˜,𝑁 + 2) if 𝑒(π‘˜ βˆ’ 1, 𝑁 + 1) = 𝑒(π‘˜,𝑁 + 1) = βˆ’1,otherwise do nothing. Finally, if the square (0, 𝑁 + 1) is black and 𝑒(βˆ’1, 𝑁 + 1) = βˆ’1 then reverse the square(βˆ’1, 𝑁 +2), and if the square (𝑀,𝑁 +1) is black and 𝑒(𝑀,𝑁 +1) = βˆ’1 then reverse the square (𝑀 +1, 𝑁 +2).The result is a field, positive in the whole rectangle (𝑀 + 1)×𝑁 .30. Answer: Zero, if one of the sides of the rectangle is 1, and any even number no greater than the number ofwhite squares in the rectangle, if both sides are greater than 1.

Hint. Denote by 𝑛(π‘₯, 𝑦, 𝑒) the number of vertices of the square (π‘₯, 𝑦) where the value of 𝑒 is negative. Noticethat any vertex that is strictly inside the rectangle (not on the boundary) is a vertex of exactly four squares inthat rectangle, two of which are white. Therefore, the sum of 𝑛(π‘₯, 𝑦, 𝑒) over all white squares in the rectangle iseven. Hence the number of negative white squares is even.

Let us show that any even number no greater than the number of white squares in the rectangle is indeedpossible.

If both the length and the width of the rectangle are even, then for any field that is positive everywhere in therectangle except for 𝑛 vertices with both coordinates even there are exactly 2𝑛 negative white squares.

For odd length or width (greater than 1) see the figure below. The identically +1 field has no negative whitesquares. The field 𝑒2𝑛+2 with 2𝑛 + 2 negative white squares is obtained from the field 𝑒2𝑛 by changing all thevalues of 𝑒2𝑛 to the opposite in any single encircled area where they have not been changed before.

31

31. Answer. Recall that 𝑒(π‘₯, 𝑦) is the value of 𝑒 in the top-right vertex of the square (π‘₯, 𝑦). Then

π‘Ž1(π‘₯, 𝑦, 𝑒) =1√2𝑒(π‘₯, 𝑦 βˆ’ 1)(π‘Ž1(π‘₯+ 1, 𝑦 βˆ’ 1, 𝑒) + π‘Ž2(π‘₯+ 1, 𝑦 βˆ’ 1, 𝑒)),

π‘Ž2(π‘₯, 𝑦, 𝑒) =1√2𝑒(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)(βˆ’π‘Ž1(π‘₯βˆ’ 1, 𝑦 βˆ’ 1, 𝑒) + π‘Ž2(π‘₯βˆ’ 1, 𝑦 βˆ’ 1, 𝑒)).

Hint. The solution of the analogue of Problem 4 is very similar to the original one, only the factors 𝑒(π‘₯, π‘¦βˆ’ 1)and 𝑒(π‘₯ βˆ’ 1, 𝑦 βˆ’ 1) are added due to the last step of the path 𝑠 passing through the top-right vertex of either(π‘₯, 𝑦 βˆ’ 1) or (π‘₯βˆ’ 1, 𝑦 βˆ’ 1).

Let us prove the analogue of Problem 5 by induction over 𝑦. The step of induction follows immediately fromthe following computation:βˆ‘

π‘₯∈Z𝑃 (π‘₯, 𝑦 + 1, 𝑒) =

βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦 + 1, 𝑒)2 + π‘Ž2(π‘₯, 𝑦 + 1, 𝑒)2

]=

βˆ‘π‘₯∈Z

π‘Ž1(π‘₯, 𝑦 + 1, 𝑒)2 +βˆ‘π‘₯∈Z

π‘Ž2(π‘₯, 𝑦 + 1, 𝑒)2 =

=1

2

βˆ‘π‘₯∈Z

𝑒(π‘₯, 𝑦)2(π‘Ž1(π‘₯+ 1, 𝑦, 𝑒) + π‘Ž2(π‘₯+ 1, 𝑦, 𝑒))2 +1

2

βˆ‘π‘₯∈Z

𝑒(π‘₯βˆ’ 1, 𝑦)2(π‘Ž2(π‘₯βˆ’ 1, 𝑦, 𝑒)βˆ’ π‘Ž1(π‘₯βˆ’ 1, 𝑦, 𝑒))2 =

=βˆ‘π‘₯∈Z

(π‘Ž1(π‘₯, 𝑦, 𝑒) + π‘Ž2(π‘₯, 𝑦, 𝑒))2

2+βˆ‘π‘₯∈Z

(π‘Ž2(π‘₯, 𝑦, 𝑒)βˆ’ π‘Ž1(π‘₯, 𝑦, 𝑒))2

2=

βˆ‘π‘₯∈Z

[π‘Ž1(π‘₯, 𝑦, 𝑒)

2 + π‘Ž2(π‘₯, 𝑦, 𝑒)2]=

βˆ‘π‘₯∈Z

𝑃 (π‘₯, 𝑦, 𝑒).

32. Answer: 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) =

⎧βŽͺβŽͺβŽͺβŽͺ⎨βŽͺβŽͺβŽͺβŽͺβŽ©π‘ƒ (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,βˆ’), if 𝐸 = (π‘₯+ 1, 𝑦), 𝐸′ = (π‘₯β€² + 1, 𝑦),

𝑃 (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,+), if 𝐸 = (π‘₯+ 1, 𝑦), 𝐸′ = (π‘₯β€² βˆ’ 1, 𝑦),

𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,βˆ’), if 𝐸 = (π‘₯βˆ’ 1, 𝑦), 𝐸′ = (π‘₯β€² + 1, 𝑦),

𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,+), if 𝐸 = (π‘₯βˆ’ 1, 𝑦), 𝐸′ = (π‘₯β€² βˆ’ 1, 𝑦).

Solution. Notice that οΏ½οΏ½(𝑠, 𝑠′) = βˆ’π‘–π‘Ž(𝑠)π‘Ž(𝑠′), where ��𝑏 denotes the product of complex numbers οΏ½οΏ½ and οΏ½οΏ½.Without loss of generality suppose that π‘₯β€² > π‘₯. Due to the condition π‘₯0 β‰₯ 2𝑦 there are no paths starting at 𝐴 andending in 𝐹 β€² and no paths starting at 𝐴′ and ending in 𝐹 .

Thereforeβˆ‘

𝑠:𝐴𝐡→𝐸′𝐹 ′𝑠′:𝐴′𝐡′→𝐸𝐹

οΏ½οΏ½(𝑠, 𝑠′) = 0 and οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) =βˆ‘

𝑠:𝐴𝐡→𝐸𝐹𝑠′:𝐴′𝐡′→𝐸′𝐹 β€²

οΏ½οΏ½(𝑠, 𝑠′).

If 𝐸 = (π‘₯βˆ’ 1, 𝑦) and 𝐸′ = (π‘₯β€² βˆ’ 1, 𝑦) then οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = βˆ’π‘–βˆ‘π‘ ,𝑠′

οΏ½οΏ½(𝑠)π‘Ž(𝑠′), where the sum is over

𝑠 and 𝑠′ ending with an upwards-right move. Therefore, οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = βˆ’π‘–π‘Ž(π‘₯, 𝑦,+)π‘Ž(π‘₯β€² βˆ’ π‘₯0, 𝑦,+).Hence 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = 𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,+).

If 𝐸 = (π‘₯+1, 𝑦) and 𝐸′ = (π‘₯β€²βˆ’ 1, 𝑦) then οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = βˆ’π‘–βˆ‘π‘ ,𝑠′

οΏ½οΏ½(𝑠)π‘Ž(𝑠′), where the sum is over 𝑠

ending with an upwards-left move and 𝑠′ ending with an upwards right move. Hence 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) =𝑃 (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,+).

If 𝐸 = (π‘₯βˆ’ 1, 𝑦) and 𝐸′ = (π‘₯β€²+1, 𝑦) then οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = βˆ’π‘–βˆ‘π‘ ,𝑠′

οΏ½οΏ½(𝑠)π‘Ž(𝑠′), where the sum is over 𝑠

ending with an upwards-right move and 𝑠′ ending with an upwards left move. Hence 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) =𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,βˆ’).

32

If 𝐸 = (π‘₯+ 1, 𝑦) and 𝐸′ = (π‘₯β€² + 1, 𝑦) then οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = βˆ’π‘–βˆ‘π‘ ,𝑠′

οΏ½οΏ½(𝑠)π‘Ž(𝑠′), where the sum is over

𝑠 and 𝑠′ ending with an upwards-left move. Hence 𝑃 (𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) = 𝑃 (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯β€² βˆ’ π‘₯0, 𝑦,βˆ’).

33. Hint. Suppose that paths 𝑠 : 𝐴𝐡 β†’ 𝐸𝐹 and 𝑠′ : 𝐴′𝐡′ β†’ 𝐸′𝐹 β€² have a common move. Suppose that 𝐢𝐷 is thefirst common move of 𝑠 and 𝑠′. Denote by 𝑠 the path that coincides with 𝑠 up to the square 𝐢 and with 𝑠′ after 𝐢.and 𝑠′ the path that coincides with 𝑠′ before 𝐢 and with 𝑠 after. The paths 𝑠, 𝑠′ start with the moves 𝐴𝐡,𝐴′𝐡′

and end with moves 𝐸′𝐹 β€² and 𝐸𝐹 respectively.The total number of turns in the paths 𝑠 and 𝑠′ equals the total number of turns in 𝑠 and 𝑠′ , because 𝐢𝐷 is a

common move for all of them. Therefore, οΏ½οΏ½(𝑠, 𝑠′) = οΏ½οΏ½(𝑠, 𝑠′). Notice that 𝑓 : (𝑠, 𝑠′) ↦→ (𝑠, 𝑠′) is a bijection, because𝐢𝐷 is exactly the first common move. One of (𝑠, 𝑠′) and (𝑠, 𝑠′) contributes to οΏ½οΏ½(𝐸𝐹,𝐸′𝐹 β€²) with the coefficient(+1) and the other one with the the coefficient (βˆ’1). Therefore, these pairs cancel each other.

34. Hint. Let us prove this equality by induction over 𝑦. The base is obvious.Note that by analogy with Problem 11, οΏ½οΏ½(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) is parallel to either the line π‘₯ = 0 or the line

𝑦 = 0. Denote by π‘Ž1(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) and π‘Ž2(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) its coordinates.Fix the starting moves 𝐴𝐡 and 𝐴′𝐡′ of the paths. Denote by π‘Ž(𝐸,𝐸′, ↑,+,βˆ’) = π‘Ž1(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²),

where 𝐸𝐹 is an upwards-right move, 𝐸′𝐹 β€² is an upwards-left move, and π‘Ž1(𝐴𝐡,𝐴′𝐡′ β†’ 𝐸𝐹,𝐸′𝐹 β€²) is the (possibly)non-zero coordinate of the corresponding vector. π‘Ž(𝐸,𝐸′, ↑,+,+), π‘Ž(𝐸,𝐸′, ↑,βˆ’,+) and π‘Ž(𝐸,𝐸′, ↑,βˆ’,βˆ’) are definedanalogously. Denote π‘Ž(𝐸,𝐸′, ↓,+,βˆ’) = π‘Ž1(𝐴𝐡,𝐴′𝐡′ β†’ 𝐷𝐸,𝐷′𝐸′), where 𝐷𝐸 is an upwards-right move and 𝐷′𝐸′

is an upwards-left move. π‘Ž(𝐸,𝐸′, ↓,+,+), π‘Ž(𝐸,𝐸′, ↓,βˆ’,+) and π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’) are defined analogously.Note that the induction step is a proof of the equalityβˆ‘

𝐸,𝐸′

((π‘Ž(𝐸,𝐸′, ↑,+,+))2 + (π‘Ž(𝐸,𝐸′, ↑,+,βˆ’))2 + (π‘Ž(𝐸,𝐸′, ↑,βˆ’,+))2 + (π‘Ž(𝐸,𝐸′, ↑,βˆ’,βˆ’))2) =

=βˆ‘πΈ,𝐸′

((π‘Ž(𝐸,𝐸′, ↓,+,+))2 + (π‘Ž(𝐸,𝐸′, ↓,+,βˆ’))2 + (π‘Ž(𝐸,𝐸′, ↓,βˆ’,+))2 + (π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’))2).

We will prove this equation by proving that for every pair (𝐸,𝐸′), where 𝐸 and 𝐸′ have the same 𝑦-coordinate,

(π‘Ž(𝐸,𝐸′, ↑,+,+))2 + (π‘Ž(𝐸,𝐸′, ↑,+,βˆ’))2 + (π‘Ž(𝐸,𝐸′, ↑,βˆ’,+))2 + (π‘Ž(𝐸,𝐸′, ↑,βˆ’,βˆ’))2 =

=(π‘Ž(𝐸,𝐸′, ↓,+,+))2 + (π‘Ž(𝐸,𝐸′, ↓,+,βˆ’))2 + (π‘Ž(𝐸,𝐸′, ↓,βˆ’,+))2 + (π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’))2.

Analogously to Problem 4 we obtain

π‘Ž(𝐸,𝐸′, ↑,+,+) =1

2(π‘Ž(𝐸,𝐸′, ↓,+,+)βˆ’ π‘Ž(𝐸,𝐸′, ↓,+,βˆ’)βˆ’ π‘Ž(𝐸,𝐸′, ↓,βˆ’,+)βˆ’ π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’)),

π‘Ž(𝐸,𝐸′, ↑,+,βˆ’) =1

2(π‘Ž(𝐸,𝐸′, ↓,+,+)βˆ’ π‘Ž(𝐸,𝐸′, ↓,+,βˆ’) + π‘Ž(𝐸,𝐸′, ↓,βˆ’,+) + π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’)),

π‘Ž(𝐸,𝐸′, ↑,βˆ’,+) =1

2(π‘Ž(𝐸,𝐸′, ↓,+,+) + π‘Ž(𝐸,𝐸′, ↓,+,βˆ’)βˆ’ π‘Ž(𝐸,𝐸′, ↓,βˆ’,+) + π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’)),

π‘Ž(𝐸,𝐸′, ↑,βˆ’,βˆ’) =1

2(βˆ’π‘Ž(𝐸,𝐸′, ↓,+,+)βˆ’ π‘Ž(𝐸,𝐸′, ↓,+,βˆ’)βˆ’ π‘Ž(𝐸,𝐸′, ↓,βˆ’,+) + π‘Ž(𝐸,𝐸′, ↓,βˆ’,βˆ’)).

With this, we can prove our equation by substitution and expansion.

35. Answer: 𝑃 (𝐴𝐡,𝐡′𝐴′ β†’ 𝐸𝐹,𝐹 ′𝐸′) =

⎧βŽͺβŽͺβŽͺβŽͺ⎨βŽͺβŽͺβŽͺβŽͺβŽ©π‘ƒ (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯0 βˆ’ π‘₯β€², 𝑦,βˆ’), if 𝐸 = (π‘₯+ 1, 𝑦), 𝐸′ = (π‘₯β€² + 1, 𝑦),

𝑃 (π‘₯, 𝑦,βˆ’)𝑃 (π‘₯0 βˆ’ π‘₯β€², 𝑦,+), if 𝐸 = (π‘₯+ 1, 𝑦), 𝐸′ = (π‘₯β€² βˆ’ 1, 𝑦),

𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯0 βˆ’ π‘₯β€², 𝑦,βˆ’), if 𝐸 = (π‘₯βˆ’ 1, 𝑦), 𝐸′ = (π‘₯β€² + 1, 𝑦),

𝑃 (π‘₯, 𝑦,+)𝑃 (π‘₯0 βˆ’ π‘₯β€², 𝑦,+), if 𝐸 = (π‘₯βˆ’ 1, 𝑦), 𝐸′ = (π‘₯β€² βˆ’ 1, 𝑦).

Hint. Notice that οΏ½οΏ½(𝑠, 𝑠′) = π‘–π‘Ž(𝑠)π‘Ž*(𝑠′), where ��𝑏 denotes the product of complex numbers οΏ½οΏ½ and οΏ½οΏ½, and οΏ½οΏ½*

denotes conjugation.For the remaining problems, we only give preliminary draft solutions.

36. Answer: for each 𝑒fin we have

𝑃 (2, 2, 𝑒fin, 1,+) = 1/4, 𝑃 (0, 2, 𝑒fin, 1,βˆ’) = 1/4,

𝑃 (3, 3, 𝑒fin, 1,+) = 1/32, 𝑃 (1, 3, 𝑒fin, 1,βˆ’) = 1/32,

𝑃 (1, 3, 𝑒fin, 1,+) = 1/32, 𝑃 (βˆ’1, 3, 𝑒fin, 1,βˆ’) = 1/32;

all the other probabilities vanish for 𝑦 = 2, 3.

37. Answer :βˆ‘π‘’οΏ½οΏ½(𝑔, 𝑠, 𝑒) = 23π‘¦βˆ’4(1 + 𝑔2)

π‘¦βˆ’12 cos((π‘Ÿ βˆ’ 𝑙) arctan(𝑔))π‘Ž(𝑠).

33

Hint. For brevity, denote 𝑍 = 1

2(π‘¦βˆ’2)2 (1+𝑔2)(π‘¦βˆ’1)2

2

.

Suppose that π‘Ÿ and 𝑙 are the numbers of white squares in 𝑅 to the right and to the left from 𝑠 respectively.Notice that π‘Ÿ and 𝑙 are odd if and only if the number of β€œβˆ’β€ signs on the path 𝑠 is odd (that is, οΏ½οΏ½(𝑠, 𝑒) = βˆ’οΏ½οΏ½(𝑠)).

Easy-to-prove lemma: if 𝑏 is the number of black squares in 𝑅 not touching the boundary, then the numberof fields 𝑒 with exactly 2π‘˜ negative white squares equals 2𝑏 times the number of ways to assign β€œβˆ’β€ to 2π‘˜ whitesquares in 𝑅. The factor 2𝑏 is the number of gauge transformations from the Problem 28.

By the lemma we have

βˆ‘π‘’

οΏ½οΏ½(𝑔, 𝑠, 𝑒) = π‘βˆ‘π‘’

(βˆ’π‘–π‘”)𝑛��(𝑠, 𝑒) = 2𝑏𝑍

⎑⎣ βˆ‘π‘˜,π‘š 𝑒𝑣𝑒𝑛

(βˆ’π‘–π‘”)π‘˜+π‘š

(π‘Ÿ

π‘˜

)(𝑙

π‘š

)βˆ’

βˆ‘π‘˜,π‘š π‘œπ‘‘π‘‘

(βˆ’π‘–π‘”)π‘˜+π‘š

(π‘Ÿ

π‘˜

)(𝑙

π‘š

)⎀⎦ οΏ½οΏ½(𝑠) =

= 2π‘βˆ’2𝑍[((1 + 𝑖𝑔)π‘Ÿ + (1βˆ’ 𝑖𝑔)π‘Ÿ)

((1 + 𝑖𝑔)𝑙 + (1βˆ’ 𝑖𝑔)𝑙

)βˆ’ ((1 + 𝑖𝑔)π‘Ÿ βˆ’ (1βˆ’ 𝑖𝑔)π‘Ÿ)

((1 + 𝑖𝑔)𝑙 βˆ’ (1βˆ’ 𝑖𝑔)𝑙

)]οΏ½οΏ½(𝑠) =

= 2𝑏𝑍𝑅𝑒((1 + 𝑖𝑔)π‘Ÿ(1βˆ’ 𝑖𝑔)𝑙

)οΏ½οΏ½(𝑠) = 2𝑏(1 + 𝑔2)

π‘Ÿ+𝑙2 𝑍 cos((π‘Ÿ βˆ’ 𝑙) arctan(𝑔))π‘Ž(𝑠).

Since 𝑏 = π‘Ÿ + 𝑙 = 𝑦2 βˆ’ 𝑦, it follows thatβˆ‘π‘’οΏ½οΏ½(𝑔, 𝑠, 𝑒) = 23π‘¦βˆ’4(1 + 𝑔2)

π‘¦βˆ’12 cos((π‘Ÿ βˆ’ 𝑙) arctan(𝑔))π‘Ž(𝑠).

38. Answer. Denote

π‘Žβˆ’βˆ’(π‘₯, 𝑦) = π‘Ž2(π‘₯, 𝑦, π‘’βˆ’, 𝑔,βˆ’), π‘Žβˆ’+(π‘₯, 𝑦) = π‘Ž1(π‘₯, 𝑦, π‘’βˆ’, 𝑔,+),

π‘Ž+βˆ’(π‘₯, 𝑦) = π‘Ž1(π‘₯, 𝑦, 𝑒+, 𝑔,βˆ’), π‘Ž++(π‘₯, 𝑦) = π‘Ž2(π‘₯, 𝑦, 𝑒+, 𝑔,+).

Denote by 𝑅(𝑦) = 𝑅 the rectangle formed by all the squares (π‘₯β€², 𝑦′) such that 1 βˆ’ 𝑦 < π‘₯β€² < 𝑦 and 0 < 𝑦′ < 𝑦.Denote 𝛼 = 𝛼(π‘₯, 𝑦) = (Δ𝑙 βˆ’ Ξ”π‘Ÿ) arctan 𝑔, where Δ𝑙 and Ξ”π‘Ÿ are the numbers of white squares in 𝑅(𝑦) outside𝑅(𝑦 βˆ’ 1) lying to the left and to the right from the square (π‘₯+ 1, 𝑦 βˆ’ 1). Denote 𝛽 = 𝛼(π‘₯βˆ’ 2, 𝑦). Then⎧βŽͺβŽͺβŽͺβŽͺβŽͺ⎨βŽͺβŽͺβŽͺβŽͺβŽͺ⎩

π‘Ž++(π‘₯, 𝑦) = cosπ›½βˆš2(π‘Ž++(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ π‘Ž+βˆ’(π‘₯βˆ’ 1, 𝑦 βˆ’ 1))βˆ’ sinπ›½βˆš

2(π‘Žβˆ’+(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ π‘Žβˆ’βˆ’(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)) ;

π‘Ž+βˆ’(π‘₯, 𝑦) = cosπ›Όβˆš2(π‘Ž++(π‘₯+ 1, 𝑦 βˆ’ 1) + π‘Ž+βˆ’(π‘₯+ 1, 𝑦 βˆ’ 1))βˆ’ sinπ›Όβˆš

2(π‘Žβˆ’+(π‘₯+ 1, 𝑦 βˆ’ 1) + π‘Žβˆ’βˆ’(π‘₯+ 1, 𝑦 βˆ’ 1)) ;

π‘Žβˆ’+(π‘₯, 𝑦) = sinπ›½βˆš2(π‘Ž++(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ π‘Ž+βˆ’(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)) + cosπ›½βˆš

2(π‘Žβˆ’+(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)βˆ’ π‘Žβˆ’βˆ’(π‘₯βˆ’ 1, 𝑦 βˆ’ 1)) ;

π‘Žβˆ’βˆ’(π‘₯, 𝑦) = sinπ›Όβˆš2(π‘Ž++(π‘₯+ 1, 𝑦 βˆ’ 1) + π‘Ž+βˆ’(π‘₯+ 1, 𝑦 βˆ’ 1)) + cosπ›Όβˆš

2(π‘Žβˆ’+(π‘₯+ 1, 𝑦 βˆ’ 1) + π‘Žβˆ’βˆ’(π‘₯+ 1, 𝑦 βˆ’ 1)) .

Hint. The latter formula defines an orthogonal transformation, which implies thatβˆ‘

π‘₯∈Z(𝑃 (π‘₯, 𝑦, 𝑔,+) +𝑃 (π‘₯, 𝑦, 𝑔,βˆ’)) = 1 by induction over 𝑦.

39. Answer. The number assigned to a basis configuration is shown below it:

1 βˆ’π‘–π‘šπœ€βˆš1+π‘š2πœ€2

1√1+π‘š2πœ€2

1√1+π‘š2πœ€2

βˆ’π‘–π‘šπœ€βˆš1+π‘š2πœ€2

βˆ’ 1.

Hint. Let us prove that the sum over all pairs of paths passing through vertices with β€œβˆ’β€ signs only equals theproduct of all the numbers assigned to the black squares in the rectangle.

Notice that the squares of type (1) do not contribute to the result at all. Squares of types (2) and (5) canonly be passed through by a path with a turn in that square, which means contributing a factor of βˆ’π‘–π‘šπœ€βˆš

1+π‘š2πœ€2to

οΏ½οΏ½(𝑠, 𝑠′). Squares of types (3) and (4) can only be passed through by a path with a turn in that square which meanscontributing a factor of 1√

1+π‘š2πœ€2to οΏ½οΏ½(𝑠, 𝑠′). Squares of type (6) can be passed through in two different ways: by

two turning or two not turning paths. When the paths do turn, the square contributes(

βˆ’π‘–π‘šπœ€βˆš1+π‘š2πœ€2

)2and when they

do not, it contributes βˆ’11+π‘š2πœ€2

, which sums up to βˆ’1.

Acknowledgements

The authors are grateful to V. Skopenkova for some of the figures, G. Chelnokov, I. Ivanov for useful discus-sions, and all participants of Summer Conference of Tournament of Towns in Arandelovac and Summer School inContemporary Mathematics in Dubna for their contribution.

34

References

[Feynman] Feynman, Richard (2006). QED: The strange theory of light and matter. Princeton University Press.ISBN 0-691-12575-9.

[Gersch] Gersch, H.A., Feynman’s relativistic chessboard as an Ising model, Int J Theor Phys 20:7 (1981), 491–501.

[Jacobson-Schulman] T Jacobson and L S Schulman, Quantum stochastics: the passage from a relativistic to anon-relativistic path integral, J Physics A 17:2 (1984), 375–383.

[Ju-Lee-Seo] Ju, H.-K.; Lee, H. & Seo, S. Integral polynomial sequences related with Krawtchouk matrices andassociated Riordan arrays Honam Math. J. 34 (2012), 297–310.

[Landafshitz] L. D. Landau, E. M. Lifshitz, L.P. Pitaevskii (1984). Electrodynamics of Continuous Media. Vol. 8(2nd ed.). Butterworth-Heinemann.

[Ord] G.N. Ord, Classical particles and the Dirac equation with an electromagnetic field, Chaos, Solitons &Fractals 8:5 (1997), 727-741.

[Ord-Gualtieri] G. N. Ord and J. A. Gualtieri, The Feynman Propagator from a Single Path, Phys. Rev. Lett. 89,250403 (2002).

35


Recommended